It seems likely to me this was driven by the `ultra` mode in 5.6, which fans subagents to do work. This mode was previously only available in the web UI (what was previously known as pro?)
It seems possible they trained this by doing full RL rollouts of agents interacting with each other. They likely view these prompts somewhat the same as raw reasoning traces, they don't want people to train directly on them.
I am unsure if this has been confirmed, but there are some signs that the opaque "compaction blob" they return from their dedicated compaction endpoint might not be text at all, rather a latent space representation of the conversation. The fact that OpenAIs compaction seems to be much higher fidelity than a lot of other providers makes me inclined to believe this.
If this is true, it doesn't seem far fetched to infer that they might be applying similar techniques to prompting subagents.
I would be curious to see if this way of spawning subagents (encrypted blob) is used when subagents of a different model type is spawned.
I think you hit the nail on the head here. Having subagent dispatch in the loop for RLVR is something we've already seen in open models, like Kimi K2.5 and later, so it's no great stretch to assume OpenAI are doing it too.
If you keep RL'ing the dispatch then the prompts are likely to keep diverging from the type of prompt a person would write (like CoT becoming increasingly incomprehensible), and that divergence is part of their competitive advantage.
> rather a latent space representation of the conversation
It's sort of insane though, you not only have dozens/hundreds of stochastic agents running on your machine, but you cannot even inspect the instructions those agents are working off of?
I've gone in to look at Claude subagent/workflows and sometimes been like "no this was a mistake to spin up" ... Codex users just get to token yolo the encrypted telephone operator instructions+shell from orchestrator to subagents?
I imagine this will be because a decent chunk of the IP in Codex is probably within its prompts, how they're built, and how they're sequenced and orchestrated, rather than in the codebase per se.
We had this discussion a few months ago where we talked about allowing people to choose an AI provider and provide their API key, thinking about enterprises with "preferred" (read: mandated) AI suppliers. We also wanted to offer the kind of very simple pricing that this is one way of enabling. But we realised pretty quickly that this would/could lead to leaking our back end prompts to customers and, although those prompts are only a part of the value add, if you could build a detailed trace of them then you'd be able to quickly reverse engineer a lot of what we're doing.
I'm unable to understand how much value can be in low-definability non deterministic prompts. It feels like the kept the right divinity spell into a chest.
I don’t disagree with your divinity spell comparison but unfortunately there is a lot of value in the prompts because these spells are the “programming languages” of LLMs.
yeah i get it too, i'm just flabbergasted that this is today's market
it reminds me of the pre-vulkan game programming days.. drivers were black boxes, game developpers had to resort to magic tricks to do stuff, until everybody got fed up and wanted some logical ground to operate
the trick about agentic systems is definitely how to do the prompting. things like automation and sandboxing are trivial in comparisson. if you generally ask via API model directly you can see what basic answers it actually yields and how fine tuning prompts and refinements to output as well as adversarial prompts etc are important to get relatively solid results.
a lot of expertise of certain domains' workflows is needed to make it functional within that domain. some of this can be yielded via prompting too etc so its also baoance of how much to prompt it vs. how much of it you wanna let it reason over itself. (if you tell it too much i lock it into a path and if you tell too little it will give incomplete results )
I don't know how you'd enforce that unless it was something you could mandate at the level of the API call, and then the API call is rejected if the customer hasn't configured it for "no transcript".
It sort of feels like an area of friction even still.
The title was fixed like 40 minutes ago, when you come back to
old browser tabs you probably want to hit that reload button before leaving a comment ;)
It's also not the first time Codex started encrypting stuff. Their excellent compaction endpoint has served up a giant encrypted blob since at least five months ago.
I've been sticking with the chat completion endpoint because of this same behavior. OAI has been subtly pushing users away from chat completion and toward the endpoints that are possible to obfuscate (responses API).
With chat completion, the reasoning process is entirely under your control. You can build a reasoning agent that uses custom MCTS techniques with GPT5.6 models today if you are willing to get your hands just a little bit dirty. You have to enable experimental flags and set options in slightly confusing ways, but it still works.
You can use models up to gpt5.5 with custom API tokens and model configuration in VS Copilot. gpt5.6 family (currently) no longer work in this setup. Presumably, because we aren't explicitly forcing reasoning_effort to none to satisfy the new moat expansion behavior.
I'm well aware of what the official propaganda states but this is simply not a fair characterization of reality.
Responses integration will lock you into OAI much more deeply than chat completion integration will. I can easily swap my inference provider right now. The business is not interested in a form of integration that is difficult to swap.
HN Title is ( edit: was ) very misleading, it makes it sound like inference is being done directly on ciphertext, which would require homomorphic encryption well advanced of what is known.
It is not misleading, quite literally what's happening is that content the agent sends sub-agents is encrypted in such a way that only OpenAIs backend can decrypt it and actually see the clear-text. Just shared this is another comment that hopefully explains things better:
> Sure. "Traditionally", your agent would send a text prompt to the sub-agent, then it goes off doing it's work. In the logs/session data, the clear-text prompt would be there, so if I want to see what's happening, I just browse the data. It's all just clear-text prompts being sent everywhere, even when you were using the experimental "sub-agents" stuff in Codex, before Sol et al was available.
> Now, when using Sol or Terra (Luna seems unaffected), instead of the agent sending clear-text prompt to the sub-agent, it sends a ciphertext generated on OpenAIs backend, which ends up being the prompt, then agent sends this ciphertext to the sub-agent, which then continues to use that for further inference to OpenAIs backend. Only delegated inter-agent messages are encrypted, not all session data. Now if you browse the data, it's all encrypted content, that can only be decrypted by OpenAI and their backend.
Edit: Re-reading, I think I understand what you mean to be misleading. You're taking "uses ciphertext for inference" quite literally, while I couldn't fit a more nuanced version within the HN title constraints. Yes, the inference at OpenAI obviously doesn't happen over the ciphertext, but from the perspective of the local user, you don't see the clear-text prompt at all, only the ciphertext.
But, please suggest alternative titles that sufficiently explain what the issue is and is more accurate, I'm sure the mods can change it once people come up with better alternatives :)
Edit2: I've updated the title from "Codex starts encrypting prompts, uses ciphertext for inference instead" to "Codex starts encrypting sub-agent prompts", hopefully it's clearer now!
Not everything is encrypted though, session data (even from the sub-agent) remains unencrypted, only select things like the prompt the (main) agent sends the sub-agent is encrypted, rest of communication between the two seems still to be plain-text.
Regardless, I've updated the title from "Codex starts encrypting prompts, uses ciphertext for inference instead" to "Codex starts encrypting sub-agent prompts", hope this makes it clearer for everyone :)
Agreed, I immediately thought that homomorphic encryption was at play here or some other kind of computation on ciphertext, given the mention of "inferencing" in the title.
I wonder if they are gonna stop us from using gpt subscriptions in alternative harnesses. If not - that doesn't matter much, codex cli is a remarkably unremarkable harness.
> I wonder if they are gonna stop us from using gpt subscriptions in alternative harnesses
Probably not, the whole app-server machinery is there to facilitate that thing, would be a huge piece to rip out of codex. This is basically the reason I end up using codex the most, as it's the easiest to integrate against, with the app-server's RPC API making it really trivial.
Besides, most of my codex usage at this point is all through custom integrations I've built using Codex's app-server, not the Codex TUI they publish. I'm sure I'm not alone in this.
But, if they suddenly start to encrypt content on our disk, so only their backends can see it, and those things are prompts and other things that are actual inputs to the inference, then who cares if it's easy to integrate against, it becomes impossible to figure out what the fuck is going on, I can't understand how the team thought this was a good idea...
Everything I do with codex is managed via Forgejo comments, issues and PRs basically. I have a tiny little Rust "conductor" that integrates with app-server and does things when issues/PRs are labeled, when I write comments on PR lines and so on, and those interactions all fire of Codex sessions that are run via Codex's app-server and lead to different outcomes.
Beats having to parse output from CLI-runs and so on. Initially this environment was running aider (which feels like years ago), was running Claude (parsing stdout) at one point but using Codex's app-server since some weeks/months back and is a lot simpler implemented now.
At least the local models they put out are pretty good for their weight class. Could be worse, could be releasing the same amount of local models as Anthropic.
Google has the benefit of billions of devices in the wild that they control. Anthropic really doesn’t have distribution for local models, makes sense they’re not playing in that space.
Oh sure, full agreement. In fact given how mid Gemini has been in cloud, this may turn out to be critical strategy to avoid losing cloud to Anthropic / OpenAI / DeepSeek / whoever, and local to Apple (doesn’t seem likely now, but that’s what BlackBerry, Creative, Intel said).
Doubt they will do it as long as Anthropic is leading in business adoption. If they become the top dog with a good lead, all bets are off. Hopefully by the time open models will be even better than gpt-5.6 sol xD.
Personally I use both, pi serves as a "personal assistant" with lots of extensions and changes made for those things specifically, and codex is for anything related to coding itself.
Quite obviously they're afraid of letting other providers see how they handle the whole multi-agent management stuff. Pretty terrible implementation though, which makes it impossible to use the multi-agent stuff as a paying user, as you have zero recourse in figuring out what went wrong, when something inevitably goes wrong.
> but also store data and sell to whoever is training
I see this as an argument against using them/Chinese models all the time, but I don't get it. I totally understand wanting to keep your data private if you're using an LLM for personal chats. But coding? I'm not working for the military, I'd gladly donate my codebase to Chinese labs if that means they can keep releasing 6-months-behind level models for 100x cheaper.
(I understand why OpenAI doesn't want this and would implement protections. I'm talking about people using this as an argument for why you as an end user shouldn't use those services.)
When you work on proprietary code with a lot of trade secrets contained in it, on a codebase that did cost millions of dollars of man-hours to build and that holds the company's IP, you tend to be very careful where you're sending that to.
Some workplace code base are legally not supposed to be shared.
More importantly, they train on not only code but also your interactions with the model, no matter how little you value your labor, there are values in it.
Yeah. I don't see the problem with Chinese prompt stealing proxies, if it's just pure free choice and discount for explicitly insecure use cases, especially when the frontier providers they route to are soft-assumed to be doing something similar.
IMO the biggest argument against "sharing" your code with LLM providers is that your approach (on a high level) will be available to your competitors on the next model release assuming they ask the right questions. Not sure how much it matters, different orgs have different risk profiles.
How do you know that they don't train their models or append your prompts to add backdoors, or compromise your supply chain by including evil dependencies? This seems hugely irresponsible.
> How do you know that they don't train their models or append your prompts to add backdoors, or compromise your supply chain by including evil dependencies?
I think most of these discussions aren't about irresponsible vibe-coders, as that whole thing is mostly a fun joke more than something serious. The rest of developers who use LLMs for development, review the code the agent writes, iterates and makes changes. Think more like pair-programming, than "Write me X then deploy to production".
I know Twitter makes it seem like everyone is doing vibe-coding and YOLOing podman images into production, but it's very uncommon in a serious/production environment to act like that. While a proper structure doesn't make it impossible for the LLMs to add backdoors either via dependencies or otherwise, but it sure makes it a lot harder.
Personally, LLMs are barely able to work alongside developers and not miss anything, I wouldn't be so worried about them being able to do normal work + malicious work at the same time, as they barely handle the first part properly yet.
> How do you know that they don't train their models or append your prompts to add backdoors, or compromise your supply chain by including evil dependencies?
Could someone explain to me where exactly the encryption is happening?
I assumed that the main agent makes calls to sub-agents locally. Does Codex work in such a way where the main agent makes calls to sub-agents in the backend (openai server) before reaching local?
Sure. "Traditionally", your agent would send a text prompt to the sub-agent, then it goes off doing it's work. In the logs/session data, the clear-text prompt would be there, so if I want to see what's happening, I just browse the data. It's all just clear-text prompts being sent everywhere, even when you were using the experimental "sub-agents" stuff in Codex, before Sol et al was available.
Now, when using Sol or Terra (Luna seems unaffected), instead of the agent sending clear-text prompt to the sub-agent, it sends a ciphertext generated on OpenAIs backend, which ends up being the prompt, then agent sends this ciphertext to the sub-agent, which then continues to use that for further inference to OpenAIs backend. Only delegated inter-agent messages are encrypted, not all session data. Now if you browse the data, it's all encrypted content, that can only be decrypted by OpenAI and their backend.
Gotcha and thank you! So the encryption is happening on the OpenAI backend and the agent's clear-text output designated to the sub-agent never reaches local.
Which is a real problem since you can't intercept/monkey patch the ciphertext to decrypt it locally to be able see the clear-text since we don't have the encryption key/algo/salt. No hacking :(
In my tests, as of this morning on the latest Codex SDK, this is not happening for sub-agent requests made in my own harness to other model providers treated as sub-agents.
Specifically, I use Opus and others for subagent execution to get alloyed properties on the workflow, and as far as I can tell, that's not affected.
Presumably, this is to hide optimizations they might be making to their own subagent processing, but that's a losing, dumb battle to fight, and misses the forest for the trees.
If we're viewing this as a _bad_ thing, I don't really see that it is any different than how Claude encrypts it's thinking. Take a peek at your ~/.claude jsonl files. You're sending thinking ciphertext back and forth to Anthropic. Presumably the thinking is either considered proprietary, or, more likely, leaks embarrassing or confidential information.
> I don't really see that it is any different than how Claude encrypts it's thinking. Take a peek at your ~/.claude jsonl files. You're sending thinking ciphertext back and forth to Anthropic.
I was already only using Claude Code to double-check if it's getting better than Codex, but with things like this, it really isn't even an alternative. What's the point of using a reasoning model if you as an end-user can't seen the reasoning? I don't think I'd be able to work like that at all, I need to have introspection into what the model is doing, and can't believe I have to say this, but also need to be able to see the plaintext of the input prompt...
I guess this implies that non-Codex harnesses get a little bit worse? In wondering what's so special about their subagents system that they feel the need to hide these messages...
Sol and Terra seems specifically post-trained to handle multi-agent orchestration, I'm guessing OpenAI feels like the trained data of when to do the spawning and what context to include for the new sub-agent is the magic in their new models, so that's what they're aiming to preserve. But, this is all a guess of course.
Right I saw them saying something along the lines of "they're good at subagents". But this seems true even with third party harnesses. So I'm wondering what Codex is hiding.
The only thing I've found impacting this, is when you specifically use the "Ultra" thinking/reasoning effort, then codex adds a small part to the system prompt to further get the model to use sub-agents. Any other reasoning/thinking effort than "Ultra" and this piece is no longer in the system prompt.
Seemingly mostly a prompting thing it seems on the surface. GPT-5.5 (and maybe even GPT-5.4) already had (experimental?) support for sub-agents, remember using it even with -spark which I think was launched together with GPT-5.4 if I remember correctly, so this whole "use sub-agents" stuff most have been part of the training data for quite some time already, but maybe they've mainly been iterating on the prompt themselves since then.
The title is a bit confusing, they're not using ciphertext for inference – they're passing ciphertext around in cases where an agent calls into another agent without exposing the plaintext to the end-user
Inference is still done in plaintext after this multi-agent message gets decrypted in the server side
Using ciphertext for inference would mean it's not a very secure ciphertext.
These two ideas don't compute for me.
Same thing with homomorphic encryption. I don't get it. If you can gain any knowledge from a ciphertext, you just found a way to exploit the ciphertext to me.
> Using ciphertext for inference would mean it's not a very secure ciphertext.
Inference is done in plain text. It's just that some parts of the response can be encrypted. While I haven't looked into this specific implementation, here's a short "how I'd do it" if I wanted to implement this:
Before:
[] - encrypted
{} - plain text
1. user -> please do this -> server
2. user <- a) [thinking1] encrypted; b) {answer1} plain text <- server
3. user -> please do this -> [thinking1] (sent encrypted as received) -> {answer1} -> good but do this instead -> server
4. user <- [...] <- [thinking2] ; {answer2}
(here the server decrypts the thinking parts, adds them to the conversation, does the inference, and sends back the new thinking trace (encrypted as well) and the new answer
After:
1. user -> please do this long task -> server
2. user <- [thinking1] ; {tool_agent_spawn([params1])} ; {answer1} (e.g. would you like me to explore or do a quick hack?) <- server
3. user -> please do this long task -> (decides if explore or message) spawn([params1]) / message -> server
3. a) if no explore -> send message as usual
3. b) if explore execute spawn that in turns begins 2 channels
4. user <- [channel_1_thinking] ; {channel_1_answer} ; [channel_2_thinking] ; {channel_2_answer} ... <- server
So the server always does inference on plain text. But it sends the "important" bits encrypted, and you only send those back if you as the user want to (or need to, or choose to, etc). The idea is that the client still gets to decide on "local" things, but the server keeps the important bits from reaching the client. In this particular case, the [params] are encrypted bits that can include prompts, etc.
The idea of homomorphic encryption is to do things without the knowledge, and not gaining the knowledge. If ciphertext contains a number, and you don't need to know what number it does to always be able to multiply it by 2, you succeeded - as a simple example.
It still just sounds like fancy obfuscation to me. I've read alot of examples trying to understand but I can't get past that being able to run processes on ciphertext in a way you can learn something doesn't make sense without me changing my definition of what I think encryption means.
To actually make it work, you would need to preformat your data very specifically, and the data you want to allow to be processed would need a subkey to unlock the parts you want processed.
I don't see a way to make it a open standard. The processing steps would need to be part of the key.
Anyway If someone figured it out I would be very interested to be sure they weren't just trying to slap a it's encrypted to meet some standards required. And..also, what amount of data processing does Alice need to do that required outsourcing to Bobs machines. Data processing and anaysis is cheap.
They’re not using ciphertext in inference. They are encrypting agent responses on their servers if it’s going to a subagent on the client. The subagent will send it back to their servers for inference.
Only their servers have the keys, so they can decrypt when running inference.
Seems fairly obvious what the point from OpenAI's side is (protect what they see as the moat, that a model is "good at spawning sub-agents"), but what's really strange to me is that the team somehow didn't manage to push back on this, it's so clearly disadvantageous to developers who are trying to rely on Codex for real work. For this we need introspection into what exactly is going on, hiding the prompts is just so backwards from what I expect from OpenAI.
> protect what they see as the moat, that a model is "good at spawning sub-agents"
Yes, that is the obvious answer. I was looking for an explanation as to why and why now. Codex is open source after all. They used to not do it. Agent prompts more generally are also not encrypted, and continue to be.
This particular change just looks unintuitive to me.
They must think they have some secret sauce they don't want others to learn. How to optimally instruct sub agents for example. If they hide the sub agent prompts, other models cannot be trained to emulate.
Oh and you can't even use local models or other providers for the sub-agents. You're locked-in.
> Is it mainly about how the main/orchestrator agent communicates with its subagents ?
Yes
> If desired the user can always see what the sub agent is doing in detail ?
Well, no, that's the problem, you're currently not allowed nor is it even possible, to see the exact prompt the main agent sent the sub agent. This is the problem.
> Isn't it the same in case of claude as well ?
No idea, but if Claude Code makes it so it's impossible to inspect what the sub-agents actually received before they started their work, then I'll say it's similarly impossible to rely on Claude Code if so.
No normative opinion on whether this is justified or not, but noting that this is only for parent -> subagent spawns/messages, and only for the `multi_agent_v2` feature (currently experimental / off by default).
Notably, subagent output is still in plaintext.
EDIT: Title was now clarified. But wanted to expand that this is actually enabled for 5.6 Ultra it appears, which does subagent orchestration more natively in the API rather than direct tool calls; they are beginning to treat subagents as similar to chain-of-thought traces (already encrypted) rather than traditional tool calls.
> and only for the `multi_agent_v2` feature (currently experimental / off by default).
Wrong, this is enabled by default for Sol and Terra (not Luna), no way of avoiding this short of patching the client yourself, and that still doesn't make the backend endpoints work, they want the ciphertext that OpenAI creates on their side.
> but noting that this is only for parent -> subagent spawns/messages
This is almost fully correct though, the encryption only seems to be for the initial prompt the main model sends the sub-agent, not all communication and not regarding the state of the sub-agent at all.
So you can inspect what the sub-agent is doing currently, and the output, but you cannot see what the initial prompt the sub-agent got started with.
Couldn't you just instruct the model to always use your tool call to spawn subagents? Subagents are not some magical thing; it's just another prompt with a couple tool calls for plumbing. One of my colleagues made his own subagent harness earlier this year before codex had them at all.
The only way these AI labs can get the app layer lock-in they need is if they can get customers used to writing them a blank check: “here, take my data and my system, do ‘stuff’ and bill me for it.”
Between this and the recent Grok upload breach, I consider these products radioactive.
Ok, so help me get this right: I ask the LLM for something, it generates prompts for sub-agents and sends them back to my client for it to call the sub-agents. Now, those sub-agent prompts are encrypted messages that the sub-agents will decrypt (by hitting a backend) to do their work.
Might as well just stuff the prompts in a database and only hand back the primary key to the client to hand off to the sub-agents. Keeps the same “data security” without the overhead of encryption (especially since encryption and decryption are happening in the same domain)
> sub-agents will decrypt (by hitting a backend) to do their work
Your local harness never decrypts the prompt, and only the OpenAI backend does. Your harness still sees tool calls in the transcript so it can act, but you lose (some) visibility as to why the subagent chooses to do so.
Imagine seeing this transcript during forensics:
[encrypted blob][thinking summary: I need to drop the prod database][shell: psql "drop database users"]
I imagine main agent tells subagent something along the lines of: use this tool on this local data with these instructions in ciphertext. Otherwise yeah, encryption would be redundant.
It's to prevent collection of queries from users that are coming from resellers/proxies, for reasons of economy or bypassing region blocks etc. The users are using the stock client and may believe they are using direct OpenAI servers.
This is very obviously a countermeasure against distillers, illicit resellers, and the like. The scale and competence of the Chinese black (grey?) market has become a serious threat that can’t be ignored.
They always talk about transparency and all but it never was as opaque as it is going on now.
There is no possible audit trail. No possible way to review what happened to validate the result. But even worse, no you will be billed somehow randomly. 20 sub agents started to do something we don't know. No way to now if it was legitimate, if it is just burning tokens or agents doing the same work on loop...
When I say things like that, I'm talking about a hypothetical version that would be computationally possible. I'm not talking about today's homomorphic encryption.
Apple's Private Cloud Compute is E2EE between the client and the attested node. Not sure if anyone else is legitimately doing that -- Apple has definitely gone the furthest in terms of verifiably ensuring that requests and responses are not misusable by Apple.
This title is easy to misinterpret. If I understand correctly: Codex now encrypts sub-agent prompts and hides those prompts from the user.
edit: originally was "Codex starts encrypting prompts, uses cyphertext for inference instead"
It seems likely to me this was driven by the `ultra` mode in 5.6, which fans subagents to do work. This mode was previously only available in the web UI (what was previously known as pro?)
It seems possible they trained this by doing full RL rollouts of agents interacting with each other. They likely view these prompts somewhat the same as raw reasoning traces, they don't want people to train directly on them.
I am unsure if this has been confirmed, but there are some signs that the opaque "compaction blob" they return from their dedicated compaction endpoint might not be text at all, rather a latent space representation of the conversation. The fact that OpenAIs compaction seems to be much higher fidelity than a lot of other providers makes me inclined to believe this.
If this is true, it doesn't seem far fetched to infer that they might be applying similar techniques to prompting subagents.
I would be curious to see if this way of spawning subagents (encrypted blob) is used when subagents of a different model type is spawned.
I think you hit the nail on the head here. Having subagent dispatch in the loop for RLVR is something we've already seen in open models, like Kimi K2.5 and later, so it's no great stretch to assume OpenAI are doing it too.
If you keep RL'ing the dispatch then the prompts are likely to keep diverging from the type of prompt a person would write (like CoT becoming increasingly incomprehensible), and that divergence is part of their competitive advantage.
> rather a latent space representation of the conversation
Student/teacher models derived from the same checkpoint convey a lot of latent information through token choice, as in: https://techxplore.com/news/2026-04-ai-chatbot-student-owls....
I wonder if this is something they can take advantage of by training on compaction inside of the RLVR loop?
It's sort of insane though, you not only have dozens/hundreds of stochastic agents running on your machine, but you cannot even inspect the instructions those agents are working off of?
I've gone in to look at Claude subagent/workflows and sometimes been like "no this was a mistake to spin up" ... Codex users just get to token yolo the encrypted telephone operator instructions+shell from orchestrator to subagents?
I imagine this will be because a decent chunk of the IP in Codex is probably within its prompts, how they're built, and how they're sequenced and orchestrated, rather than in the codebase per se.
We had this discussion a few months ago where we talked about allowing people to choose an AI provider and provide their API key, thinking about enterprises with "preferred" (read: mandated) AI suppliers. We also wanted to offer the kind of very simple pricing that this is one way of enabling. But we realised pretty quickly that this would/could lead to leaking our back end prompts to customers and, although those prompts are only a part of the value add, if you could build a detailed trace of them then you'd be able to quickly reverse engineer a lot of what we're doing.
So we quickly dropped that idea.
I'm unable to understand how much value can be in low-definability non deterministic prompts. It feels like the kept the right divinity spell into a chest.
I don’t disagree with your divinity spell comparison but unfortunately there is a lot of value in the prompts because these spells are the “programming languages” of LLMs.
yeah i get it too, i'm just flabbergasted that this is today's market
it reminds me of the pre-vulkan game programming days.. drivers were black boxes, game developpers had to resort to magic tricks to do stuff, until everybody got fed up and wanted some logical ground to operate
> understand how much value can be in low-definability non deterministic prompts
When was the last time you used an LLM to evaluate how true those last part(s) still are?
I also love how you went from "I'm unable to understand" to "This is surely right", it's a good representation of the software ecosystem at large :)
the trick about agentic systems is definitely how to do the prompting. things like automation and sandboxing are trivial in comparisson. if you generally ask via API model directly you can see what basic answers it actually yields and how fine tuning prompts and refinements to output as well as adversarial prompts etc are important to get relatively solid results.
a lot of expertise of certain domains' workflows is needed to make it functional within that domain. some of this can be yielded via prompting too etc so its also baoance of how much to prompt it vs. how much of it you wanna let it reason over itself. (if you tell it too much i lock it into a path and if you tell too little it will give incomplete results )
Perhaps AI providers should support this natively: the customer supplies the API key but doesn't get access to the transcripts.
I don't know how you'd enforce that unless it was something you could mandate at the level of the API call, and then the API call is rejected if the customer hasn't configured it for "no transcript".
It sort of feels like an area of friction even still.
Yeah I thought "wow, some homomorphic encryption* stuff", but then "nah, usual greed".
* https://en.wikipedia.org/wiki/Homomorphic_encryption
The title was fixed like 40 minutes ago, when you come back to old browser tabs you probably want to hit that reload button before leaving a comment ;)
It's also not the first time Codex started encrypting stuff. Their excellent compaction endpoint has served up a giant encrypted blob since at least five months ago.
Feels strongly like we're in the late-stage-AI-unicorn phase. If this is really their moat then the Chinese companies will win.
I've been sticking with the chat completion endpoint because of this same behavior. OAI has been subtly pushing users away from chat completion and toward the endpoints that are possible to obfuscate (responses API).
With chat completion, the reasoning process is entirely under your control. You can build a reasoning agent that uses custom MCTS techniques with GPT5.6 models today if you are willing to get your hands just a little bit dirty. You have to enable experimental flags and set options in slightly confusing ways, but it still works.
You can use models up to gpt5.5 with custom API tokens and model configuration in VS Copilot. gpt5.6 family (currently) no longer work in this setup. Presumably, because we aren't explicitly forcing reasoning_effort to none to satisfy the new moat expansion behavior.
responses api provided many benefits over chat completions https://developers.openai.com/api/docs/guides/migrate-to-res.... Any reasoning model which is all the new models these days should use responses as per the recommendation
I'm well aware of what the official propaganda states but this is simply not a fair characterization of reality.
Responses integration will lock you into OAI much more deeply than chat completion integration will. I can easily swap my inference provider right now. The business is not interested in a form of integration that is difficult to swap.
HN Title is ( edit: was ) very misleading, it makes it sound like inference is being done directly on ciphertext, which would require homomorphic encryption well advanced of what is known.
It is not misleading, quite literally what's happening is that content the agent sends sub-agents is encrypted in such a way that only OpenAIs backend can decrypt it and actually see the clear-text. Just shared this is another comment that hopefully explains things better:
> Sure. "Traditionally", your agent would send a text prompt to the sub-agent, then it goes off doing it's work. In the logs/session data, the clear-text prompt would be there, so if I want to see what's happening, I just browse the data. It's all just clear-text prompts being sent everywhere, even when you were using the experimental "sub-agents" stuff in Codex, before Sol et al was available.
> Now, when using Sol or Terra (Luna seems unaffected), instead of the agent sending clear-text prompt to the sub-agent, it sends a ciphertext generated on OpenAIs backend, which ends up being the prompt, then agent sends this ciphertext to the sub-agent, which then continues to use that for further inference to OpenAIs backend. Only delegated inter-agent messages are encrypted, not all session data. Now if you browse the data, it's all encrypted content, that can only be decrypted by OpenAI and their backend.
Edit: Re-reading, I think I understand what you mean to be misleading. You're taking "uses ciphertext for inference" quite literally, while I couldn't fit a more nuanced version within the HN title constraints. Yes, the inference at OpenAI obviously doesn't happen over the ciphertext, but from the perspective of the local user, you don't see the clear-text prompt at all, only the ciphertext.
But, please suggest alternative titles that sufficiently explain what the issue is and is more accurate, I'm sure the mods can change it once people come up with better alternatives :)
Edit2: I've updated the title from "Codex starts encrypting prompts, uses ciphertext for inference instead" to "Codex starts encrypting sub-agent prompts", hopefully it's clearer now!
I would change
"Codex starts encrypting prompts, uses ciphertext for inference instead"
to just
"Codex starts encrypting prompts"
That is enough.
Maybe you could say sub agent prompts. The article can say the rest.
Not everything is encrypted though, session data (even from the sub-agent) remains unencrypted, only select things like the prompt the (main) agent sends the sub-agent is encrypted, rest of communication between the two seems still to be plain-text.
Regardless, I've updated the title from "Codex starts encrypting prompts, uses ciphertext for inference instead" to "Codex starts encrypting sub-agent prompts", hope this makes it clearer for everyone :)
Seconded can we change pls.
Agreed, I immediately thought that homomorphic encryption was at play here or some other kind of computation on ciphertext, given the mention of "inferencing" in the title.
My bad, fixed now, please do refresh and try with latest updated IE if you still don't see the changes.
> with latest updated IE
Internet Explorer?
I wonder if they are gonna stop us from using gpt subscriptions in alternative harnesses. If not - that doesn't matter much, codex cli is a remarkably unremarkable harness.
> I wonder if they are gonna stop us from using gpt subscriptions in alternative harnesses
Probably not, the whole app-server machinery is there to facilitate that thing, would be a huge piece to rip out of codex. This is basically the reason I end up using codex the most, as it's the easiest to integrate against, with the app-server's RPC API making it really trivial.
Besides, most of my codex usage at this point is all through custom integrations I've built using Codex's app-server, not the Codex TUI they publish. I'm sure I'm not alone in this.
But, if they suddenly start to encrypt content on our disk, so only their backends can see it, and those things are prompts and other things that are actual inputs to the inference, then who cares if it's easy to integrate against, it becomes impossible to figure out what the fuck is going on, I can't understand how the team thought this was a good idea...
What are some of the things you’re doing with the Codex app-server?
Everything I do with codex is managed via Forgejo comments, issues and PRs basically. I have a tiny little Rust "conductor" that integrates with app-server and does things when issues/PRs are labeled, when I write comments on PR lines and so on, and those interactions all fire of Codex sessions that are run via Codex's app-server and lead to different outcomes.
Beats having to parse output from CLI-runs and so on. Initially this environment was running aider (which feels like years ago), was running Claude (parsing stdout) at one point but using Codex's app-server since some weeks/months back and is a lot simpler implemented now.
Anthropic and Google already charge extra to use your own harness. That's 100% of the reason I'm using OpenAI.
If they go down that path I'll just go back to my old buddy Claude, or maybe buy a second Spark and keep it local.
Well, my backup plan is GLM. Cheap and not that bad really.
Google is really not distinguishing itself. Even the hosted inference sucks.
At least the local models they put out are pretty good for their weight class. Could be worse, could be releasing the same amount of local models as Anthropic.
Google has the benefit of billions of devices in the wild that they control. Anthropic really doesn’t have distribution for local models, makes sense they’re not playing in that space.
Sure, lots of differences. Point stands, they're distinguishing themselves in that way at least.
Oh sure, full agreement. In fact given how mid Gemini has been in cloud, this may turn out to be critical strategy to avoid losing cloud to Anthropic / OpenAI / DeepSeek / whoever, and local to Apple (doesn’t seem likely now, but that’s what BlackBerry, Creative, Intel said).
Doubt they will do it as long as Anthropic is leading in business adoption. If they become the top dog with a good lead, all bets are off. Hopefully by the time open models will be even better than gpt-5.6 sol xD.
It could also be the case that by the time business adoption picks up a lot they might not be as compute constrained. Depends on rate of growth.
Given that codex itself now ships a proxy that wraps the subscription, it seems unlikely
https://github.com/openai/codex/blob/main/codex-rs/responses...
Codex CLI not having a rewind makes it useless to me.
It has a similar function.
"remarkably unremarkable harness" is why I like it so much.
I don't, feel better with Pi with a custom set of extensions.
Personally I use both, pi serves as a "personal assistant" with lots of extensions and changes made for those things specifically, and codex is for anything related to coding itself.
A way to work around that would be to not use subagents, but delegate to other full agent invocations.
I assume this is mostly to frustrate efforts to proxy large numbers of user requests and responses and use it to train competitor models.
Quite obviously they're afraid of letting other providers see how they handle the whole multi-agent management stuff. Pretty terrible implementation though, which makes it impossible to use the multi-agent stuff as a paying user, as you have zero recourse in figuring out what went wrong, when something inevitably goes wrong.
Ah I was wondering why the Chinese black market resellers stopped working yesterday, I guess that's it
This is the reason I think. These black markets not only pool and resell subs, but also store data and sell to whoever is training.
Encryption is useful to at least stop the latter.
Ultimately same purpose as a\ ‘s trick exposed earlier, but a much nicer implementation.
> but also store data and sell to whoever is training
I see this as an argument against using them/Chinese models all the time, but I don't get it. I totally understand wanting to keep your data private if you're using an LLM for personal chats. But coding? I'm not working for the military, I'd gladly donate my codebase to Chinese labs if that means they can keep releasing 6-months-behind level models for 100x cheaper.
(I understand why OpenAI doesn't want this and would implement protections. I'm talking about people using this as an argument for why you as an end user shouldn't use those services.)
When you work on proprietary code with a lot of trade secrets contained in it, on a codebase that did cost millions of dollars of man-hours to build and that holds the company's IP, you tend to be very careful where you're sending that to.
Why would anyone working on such code send it anywhere (other than, say, to AWS for hosting)?
Source: I work on such code. We don't allow devs to use (cloud-based) LLMs.
Some workplace code base are legally not supposed to be shared.
More importantly, they train on not only code but also your interactions with the model, no matter how little you value your labor, there are values in it.
Yeah. I don't see the problem with Chinese prompt stealing proxies, if it's just pure free choice and discount for explicitly insecure use cases, especially when the frontier providers they route to are soft-assumed to be doing something similar.
IMO the biggest argument against "sharing" your code with LLM providers is that your approach (on a high level) will be available to your competitors on the next model release assuming they ask the right questions. Not sure how much it matters, different orgs have different risk profiles.
How do you know that they don't train their models or append your prompts to add backdoors, or compromise your supply chain by including evil dependencies? This seems hugely irresponsible.
> How do you know that they don't train their models or append your prompts to add backdoors, or compromise your supply chain by including evil dependencies?
I think most of these discussions aren't about irresponsible vibe-coders, as that whole thing is mostly a fun joke more than something serious. The rest of developers who use LLMs for development, review the code the agent writes, iterates and makes changes. Think more like pair-programming, than "Write me X then deploy to production".
I know Twitter makes it seem like everyone is doing vibe-coding and YOLOing podman images into production, but it's very uncommon in a serious/production environment to act like that. While a proper structure doesn't make it impossible for the LLMs to add backdoors either via dependencies or otherwise, but it sure makes it a lot harder.
Personally, LLMs are barely able to work alongside developers and not miss anything, I wouldn't be so worried about them being able to do normal work + malicious work at the same time, as they barely handle the first part properly yet.
> How do you know that they don't train their models or append your prompts to add backdoors, or compromise your supply chain by including evil dependencies?
I read the code.
Interesting. Got a link for this?
Could someone explain to me where exactly the encryption is happening?
I assumed that the main agent makes calls to sub-agents locally. Does Codex work in such a way where the main agent makes calls to sub-agents in the backend (openai server) before reaching local?
Sure. "Traditionally", your agent would send a text prompt to the sub-agent, then it goes off doing it's work. In the logs/session data, the clear-text prompt would be there, so if I want to see what's happening, I just browse the data. It's all just clear-text prompts being sent everywhere, even when you were using the experimental "sub-agents" stuff in Codex, before Sol et al was available.
Now, when using Sol or Terra (Luna seems unaffected), instead of the agent sending clear-text prompt to the sub-agent, it sends a ciphertext generated on OpenAIs backend, which ends up being the prompt, then agent sends this ciphertext to the sub-agent, which then continues to use that for further inference to OpenAIs backend. Only delegated inter-agent messages are encrypted, not all session data. Now if you browse the data, it's all encrypted content, that can only be decrypted by OpenAI and their backend.
Gotcha and thank you! So the encryption is happening on the OpenAI backend and the agent's clear-text output designated to the sub-agent never reaches local.
Which is a real problem since you can't intercept/monkey patch the ciphertext to decrypt it locally to be able see the clear-text since we don't have the encryption key/algo/salt. No hacking :(
In my tests, as of this morning on the latest Codex SDK, this is not happening for sub-agent requests made in my own harness to other model providers treated as sub-agents.
Specifically, I use Opus and others for subagent execution to get alloyed properties on the workflow, and as far as I can tell, that's not affected.
Presumably, this is to hide optimizations they might be making to their own subagent processing, but that's a losing, dumb battle to fight, and misses the forest for the trees.
There was a recent report on twitter of a GPT 5.6 sub-agent accidentally deleting the user's home directory.
I wonder if there was any safeguard failure due to loss of visibility into what the sub-agent was trying to do?
https://x.com/mattshumer_/status/2076794038456385546?s=20
Every day I get closer to only using pi
If we're viewing this as a _bad_ thing, I don't really see that it is any different than how Claude encrypts it's thinking. Take a peek at your ~/.claude jsonl files. You're sending thinking ciphertext back and forth to Anthropic. Presumably the thinking is either considered proprietary, or, more likely, leaks embarrassing or confidential information.
> I don't really see that it is any different than how Claude encrypts it's thinking. Take a peek at your ~/.claude jsonl files. You're sending thinking ciphertext back and forth to Anthropic.
I was already only using Claude Code to double-check if it's getting better than Codex, but with things like this, it really isn't even an alternative. What's the point of using a reasoning model if you as an end-user can't seen the reasoning? I don't think I'd be able to work like that at all, I need to have introspection into what the model is doing, and can't believe I have to say this, but also need to be able to see the plaintext of the input prompt...
GPT 5.6 is like Fable minus Opus plus down syndrome.
Claude "encrypting" its thinking is equally bad. Biggest IP thieves in the world worried about IP theft lmao.
At least Anthropic doesn't pretend that they have open source software in the form of Claude Code.
They get to pretend to be on the best side of every position simultaneously.
They're only encrypting thinking because AI is so dangerous and only they can be trusted to be in control of AGI.
This happens to align with lining their pockets as well.
I guess this implies that non-Codex harnesses get a little bit worse? In wondering what's so special about their subagents system that they feel the need to hide these messages...
Sol and Terra seems specifically post-trained to handle multi-agent orchestration, I'm guessing OpenAI feels like the trained data of when to do the spawning and what context to include for the new sub-agent is the magic in their new models, so that's what they're aiming to preserve. But, this is all a guess of course.
Right I saw them saying something along the lines of "they're good at subagents". But this seems true even with third party harnesses. So I'm wondering what Codex is hiding.
The only thing I've found impacting this, is when you specifically use the "Ultra" thinking/reasoning effort, then codex adds a small part to the system prompt to further get the model to use sub-agents. Any other reasoning/thinking effort than "Ultra" and this piece is no longer in the system prompt.
Seemingly mostly a prompting thing it seems on the surface. GPT-5.5 (and maybe even GPT-5.4) already had (experimental?) support for sub-agents, remember using it even with -spark which I think was launched together with GPT-5.4 if I remember correctly, so this whole "use sub-agents" stuff most have been part of the training data for quite some time already, but maybe they've mainly been iterating on the prompt themselves since then.
The black box is getting darker…I’m the end I think they would like to sell you a black box/appliance
How does this affect local models? Will all the features of Codex still work with local models?
The title is a bit confusing, they're not using ciphertext for inference – they're passing ciphertext around in cases where an agent calls into another agent without exposing the plaintext to the end-user
Inference is still done in plaintext after this multi-agent message gets decrypted in the server side
Using ciphertext for inference would mean it's not a very secure ciphertext.
These two ideas don't compute for me.
Same thing with homomorphic encryption. I don't get it. If you can gain any knowledge from a ciphertext, you just found a way to exploit the ciphertext to me.
> Using ciphertext for inference would mean it's not a very secure ciphertext.
Inference is done in plain text. It's just that some parts of the response can be encrypted. While I haven't looked into this specific implementation, here's a short "how I'd do it" if I wanted to implement this:
Before:
[] - encrypted {} - plain text
1. user -> please do this -> server
2. user <- a) [thinking1] encrypted; b) {answer1} plain text <- server
3. user -> please do this -> [thinking1] (sent encrypted as received) -> {answer1} -> good but do this instead -> server
4. user <- [...] <- [thinking2] ; {answer2}
(here the server decrypts the thinking parts, adds them to the conversation, does the inference, and sends back the new thinking trace (encrypted as well) and the new answer
After:
1. user -> please do this long task -> server
2. user <- [thinking1] ; {tool_agent_spawn([params1])} ; {answer1} (e.g. would you like me to explore or do a quick hack?) <- server
3. user -> please do this long task -> (decides if explore or message) spawn([params1]) / message -> server
3. a) if no explore -> send message as usual 3. b) if explore execute spawn that in turns begins 2 channels
4. user <- [channel_1_thinking] ; {channel_1_answer} ; [channel_2_thinking] ; {channel_2_answer} ... <- server
So the server always does inference on plain text. But it sends the "important" bits encrypted, and you only send those back if you as the user want to (or need to, or choose to, etc). The idea is that the client still gets to decide on "local" things, but the server keeps the important bits from reaching the client. In this particular case, the [params] are encrypted bits that can include prompts, etc.
The idea of homomorphic encryption is to do things without the knowledge, and not gaining the knowledge. If ciphertext contains a number, and you don't need to know what number it does to always be able to multiply it by 2, you succeeded - as a simple example.
It still just sounds like fancy obfuscation to me. I've read alot of examples trying to understand but I can't get past that being able to run processes on ciphertext in a way you can learn something doesn't make sense without me changing my definition of what I think encryption means.
I thought though the idea is that you cannot learn anything from the ciphertext, how it is processed, or what the final result is?
Unless you are a participant of the computation and you have the key, that is.
To actually make it work, you would need to preformat your data very specifically, and the data you want to allow to be processed would need a subkey to unlock the parts you want processed.
I don't see a way to make it a open standard. The processing steps would need to be part of the key.
Anyway If someone figured it out I would be very interested to be sure they weren't just trying to slap a it's encrypted to meet some standards required. And..also, what amount of data processing does Alice need to do that required outsourcing to Bobs machines. Data processing and anaysis is cheap.
They’re not using ciphertext in inference. They are encrypting agent responses on their servers if it’s going to a subagent on the client. The subagent will send it back to their servers for inference. Only their servers have the keys, so they can decrypt when running inference.
What's the idea here? Why does this seem important to OpenAI?
Seems fairly obvious what the point from OpenAI's side is (protect what they see as the moat, that a model is "good at spawning sub-agents"), but what's really strange to me is that the team somehow didn't manage to push back on this, it's so clearly disadvantageous to developers who are trying to rely on Codex for real work. For this we need introspection into what exactly is going on, hiding the prompts is just so backwards from what I expect from OpenAI.
> protect what they see as the moat, that a model is "good at spawning sub-agents"
Yes, that is the obvious answer. I was looking for an explanation as to why and why now. Codex is open source after all. They used to not do it. Agent prompts more generally are also not encrypted, and continue to be.
This particular change just looks unintuitive to me.
They must think they have some secret sauce they don't want others to learn. How to optimally instruct sub agents for example. If they hide the sub agent prompts, other models cannot be trained to emulate.
Oh and you can't even use local models or other providers for the sub-agents. You're locked-in.
Can you do a subagent by just wiring it call your own CLI script?
first they encrypt their prompts, next they'll develop their own slang and we'll need a translator just to debug our own codebase
Is it mainly about how the main/orchestrator agent communicates with its subagents ?
If desired the user can always see what the sub agent is doing in detail ?
Isn't it the same in case of claude as well ?
> Is it mainly about how the main/orchestrator agent communicates with its subagents ?
Yes
> If desired the user can always see what the sub agent is doing in detail ?
Well, no, that's the problem, you're currently not allowed nor is it even possible, to see the exact prompt the main agent sent the sub agent. This is the problem.
> Isn't it the same in case of claude as well ?
No idea, but if Claude Code makes it so it's impossible to inspect what the sub-agents actually received before they started their work, then I'll say it's similarly impossible to rely on Claude Code if so.
No normative opinion on whether this is justified or not, but noting that this is only for parent -> subagent spawns/messages, and only for the `multi_agent_v2` feature (currently experimental / off by default).
Notably, subagent output is still in plaintext.
EDIT: Title was now clarified. But wanted to expand that this is actually enabled for 5.6 Ultra it appears, which does subagent orchestration more natively in the API rather than direct tool calls; they are beginning to treat subagents as similar to chain-of-thought traces (already encrypted) rather than traditional tool calls.
> and only for the `multi_agent_v2` feature (currently experimental / off by default).
Wrong, this is enabled by default for Sol and Terra (not Luna), no way of avoiding this short of patching the client yourself, and that still doesn't make the backend endpoints work, they want the ciphertext that OpenAI creates on their side.
> but noting that this is only for parent -> subagent spawns/messages
This is almost fully correct though, the encryption only seems to be for the initial prompt the main model sends the sub-agent, not all communication and not regarding the state of the sub-agent at all.
So you can inspect what the sub-agent is doing currently, and the output, but you cannot see what the initial prompt the sub-agent got started with.
Couldn't you just instruct the model to always use your tool call to spawn subagents? Subagents are not some magical thing; it's just another prompt with a couple tool calls for plumbing. One of my colleagues made his own subagent harness earlier this year before codex had them at all.
Outrageous yet predictable.
The only way these AI labs can get the app layer lock-in they need is if they can get customers used to writing them a blank check: “here, take my data and my system, do ‘stuff’ and bill me for it.”
Between this and the recent Grok upload breach, I consider these products radioactive.
Better get ready to code manually again ;)
“Starts”? How’s this not already a TLS connection?
The prompts are now encrypted, not just the transit connections...
Ok, so help me get this right: I ask the LLM for something, it generates prompts for sub-agents and sends them back to my client for it to call the sub-agents. Now, those sub-agent prompts are encrypted messages that the sub-agents will decrypt (by hitting a backend) to do their work.
Might as well just stuff the prompts in a database and only hand back the primary key to the client to hand off to the sub-agents. Keeps the same “data security” without the overhead of encryption (especially since encryption and decryption are happening in the same domain)
> sub-agents will decrypt (by hitting a backend) to do their work
Your local harness never decrypts the prompt, and only the OpenAI backend does. Your harness still sees tool calls in the transcript so it can act, but you lose (some) visibility as to why the subagent chooses to do so.
Imagine seeing this transcript during forensics:
[encrypted blob][thinking summary: I need to drop the prod database][shell: psql "drop database users"]
I imagine main agent tells subagent something along the lines of: use this tool on this local data with these instructions in ciphertext. Otherwise yeah, encryption would be redundant.
Then why to even keep codex open source?
It's to prevent collection of queries from users that are coming from resellers/proxies, for reasons of economy or bypassing region blocks etc. The users are using the stock client and may believe they are using direct OpenAI servers.
what does that have to do with anything?
This is very obviously a countermeasure against distillers, illicit resellers, and the like. The scale and competence of the Chinese black (grey?) market has become a serious threat that can’t be ignored.
They always talk about transparency and all but it never was as opaque as it is going on now.
There is no possible audit trail. No possible way to review what happened to validate the result. But even worse, no you will be billed somehow randomly. 20 sub agents started to do something we don't know. No way to now if it was legitimate, if it is just burning tokens or agents doing the same work on loop...
If I were to guess this is to stop distilling and all of those blackmarket resellers.
Yeah nope. Full transparency only with my data.
But codex is opensource, no?
~~Finally someone doing it correctly. Love this change.~~
Edit: F really misunderstood the change, the title is misleading AF. I should have read the post before commenting lmao.
Absolutely hate it, now I guess... sigh..
Incase the title gets changed it used to say, "Codex starts encrypting prompts, uses ciphertext for inference instead"
How so?
I assume OP interpreted it as encryption that hides the prompts from OpenAI rather than OpenAI hiding information from users.
I'd be all for homomorphic encryption on inference, but as you say, this is probably mostly to prevent end users from observing intermediate results.
Homomorphic encryption for LLMs is extremely expensive and nowhere near computationally possible for the scale of current LLMs.
When I say things like that, I'm talking about a hypothetical version that would be computationally possible. I'm not talking about today's homomorphic encryption.
ooooooof yeah totally misinterpreted it lmao
Given what we know about the company, I find it crazy anyone would misinterpret along those lines lol.
to be fair, the title is very misleading, it took me a minute to understand what they meant
Apple's Private Cloud Compute is E2EE between the client and the attested node. Not sure if anyone else is legitimately doing that -- Apple has definitely gone the furthest in terms of verifiably ensuring that requests and responses are not misusable by Apple.