7 ms·

Looks like all three breaking changes are patches for inadvertent chain of thought disclosure. Someone found out (don't have the tweet handy) that if you create

by mlaux 16d ago

Looks like all three breaking changes are patches for inadvertent chain of thought disclosure. Someone found out (don't have the tweet handy) that if you created a bogus "think_deeply" tool and then forced the model to use it, it would output what is believed to be its raw thinking there - I believe the first breaking change stops this. The second two are aimed at people getting Haiku to repeat thinking blocks from other models verbatim (since it can see the decrypted version). I get that in their eyes it's an "exploit" but still kinda disappointing that they patched this

mmaunder 15d ago
Making clear the scale of distillation they’re combating.
- awestroke 15d ago
  It's only theft when people pay Anthropic for inference in order to improve their own datasets. It's not theft when Anthropic grabbed basically all ebooks and web content on the internet to build their own dataset, without paying anything to anyone
bandrami 15d ago
Aren't the "thinking" chains always just reconstructed anyways? It would be like using a debugger that just looks at the source code rather than the actual binary.
nullbio 15d ago
So basically, Anthropic can charge you for tokens you don't even see. "Trust me bro, you really did use $5,000 worth of tokens to generate that pelican". We need AI consumer rights, urgently.
- pbasista 15d ago
  I think that it would be sufficient to realize that no one is forced to use a specific AI model or model harness that has anti-consumer features built in. People should just walk away if they see something like that. This is where the actual, practical, consumer rights start. Not with the regulation. But with the customers being determined to stand up for themselves. And not just fold. There are plenty of good enough models that are open weight and whose use comes with almost no strings attached. And plenty of great harnesses such as various flavours of Pi.
  - nullbio 15d ago
    Agreed, and that's exactly why I don't use Claude anymore and haven't for months now. But I'm not most people. Most people don't vote with their wallet, unfortunately.
albert_e 15d ago
Anthropic: AI models should be explainable so that we can ensure and verify alignment. Responsible AI 101. Also Anthropic: No not like that.
giancarlostoro 16d ago
To be fair, I assume they want to hide that not from their customers, but adversaries who use the way Claude models think and reason to refine their own models.
- oofbey 16d ago
  That's their motivation, for sure. But it's also unambiguously making their product worse and harder to use legitimately. Which pushes customers further towards use of open weight models which don't have these restrictions. I don't think this is a fight they're going to win. It's also hard to have sympathy for them - they want to protect their IP, sure. But their IP was built on a corpus of dubious legal provenance. And even if the courts decide their training data are legal, most of the authors of the data would disagree. There was no consent given. I think LLM's are great - don't get me wrong. I'm glad they were built the way they were, because it's unlocking an amazing new world. But I just don't have sympathy for the "I stole this and now it's mine so you can't steal it" argument behind concealing reasoning traces.
- appplication 16d ago
  I have a hard time believing whatever prompts get Claude to reason can stay relevant secret sauce for long anyways. It’s not hard to A/B test something that gets you close enough, and it’s not Ike anthropic has uncovered the global optima of reasoning prompts.
- verdverm 16d ago
  I don't really want the models I use learning from Claude at this point. Open weight models of similar scale are available now too, so I expect this "distillation"/"stealing" chatter to wind down.
  - hyperpape 16d ago
    Maybe, but that's sort of begging the question that those open weight models aren't significantly trained using "distillation"[0] [0] not technically distillation. https://thomasdullien.github.io/posts/2026-06-15-rl-economics-morally-charged-terms-and-distillation/ https://thomasdullien.github.io/posts/2026-06-15-rl-economic...
trueno 16d ago
[dead]
sippeangelo 16d ago
These draconian "Preserved Thinking" measures they're taking are going to be an absolute pain in the ass. This alone is enough for me to move our API use off their platform entirely. It's a HUGE breaking change that they're trying to dampen by having it not affecting current customers until "in the future", see: https://platform.claude.com/docs/en/build-with-claude/preserved-thinking https://platform.claude.com/docs/en/build-with-claude/preser... You're no longer allowed to edit the context anywhere! The whole context is to become append-only, says Anthropic. No more editing the system prompt as the conversation progresses, no more dynamic loading of custom tool calling formats. Everything has to go through their built-in tools API and you aren't allowed to mess with anything in the context if it has any thinking blocks following it. This is the most intrusive "model DRM" we've seen so far!
- solenoid0937 15d ago
  You think moving will help you? OpenAI is going to do exactly the same thing soon. Unless you're moving off the frontier entirely, that is.
- albert_e 15d ago
  How do they still purport to champion alignment and explainablity of AI if the reasoning traces are going to be hidden.
- epolanski 16d ago
  Interesting, I liked to experiment with a second model "simplifying" and summarizing the previous messages and continue. Needless to say, it improved output on following messages by whatever metric I cared for. Not sure why would they prevent it. I give you a chain of messages, what do you care for what the origin is?
- l1n 16d ago
  > No more editing the system prompt as the conversation progresses, no more dynamic loading of custom tool calling formats. Hm, aiui you can support both of these via mid-conversation system turns https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages https://platform.claude.com/docs/en/build-with-claude/mid-co... - and in general you'd want to to preserve the cache and recency of the instruction anyways rather than frankensteining an off-distribution transcript. Not sure though.

by mlaux 16d ago

mmaunder 15d ago

Making clear the scale of distillation they’re combating.

awestroke 15d ago
It's only theft when people pay Anthropic for inference in order to improve their own datasets. It's not theft when Anthropic grabbed basically all ebooks and web content on the internet to build their own dataset, without paying anything to anyone

bandrami 15d ago

Aren't the "thinking" chains always just reconstructed anyways? It would be like using a debugger that just looks at the source code rather than the actual binary.

nullbio 15d ago

So basically, Anthropic can charge you for tokens you don't even see. "Trust me bro, you really did use $5,000 worth of tokens to generate that pelican". We need AI consumer rights, urgently.

pbasista 15d ago
I think that it would be sufficient to realize that no one is forced to use a specific AI model or model harness that has anti-consumer features built in. People should just walk away if they see something like that. This is where the actual, practical, consumer rights start. Not with the regulation. But with the customers being determined to stand up for themselves. And not just fold. There are plenty of good enough models that are open weight and whose use comes with almost no strings attached. And plenty of great harnesses such as various flavours of Pi.
- nullbio 15d ago
  Agreed, and that's exactly why I don't use Claude anymore and haven't for months now. But I'm not most people. Most people don't vote with their wallet, unfortunately.

albert_e 15d ago

Anthropic: AI models should be explainable so that we can ensure and verify alignment. Responsible AI 101. Also Anthropic: No not like that.

giancarlostoro 16d ago

To be fair, I assume they want to hide that not from their customers, but adversaries who use the way Claude models think and reason to refine their own models.

oofbey 16d ago
That's their motivation, for sure. But it's also unambiguously making their product worse and harder to use legitimately. Which pushes customers further towards use of open weight models which don't have these restrictions. I don't think this is a fight they're going to win. It's also hard to have sympathy for them - they want to protect their IP, sure. But their IP was built on a corpus of dubious legal provenance. And even if the courts decide their training data are legal, most of the authors of the data would disagree. There was no consent given. I think LLM's are great - don't get me wrong. I'm glad they were built the way they were, because it's unlocking an amazing new world. But I just don't have sympathy for the "I stole this and now it's mine so you can't steal it" argument behind concealing reasoning traces.
appplication 16d ago
I have a hard time believing whatever prompts get Claude to reason can stay relevant secret sauce for long anyways. It’s not hard to A/B test something that gets you close enough, and it’s not Ike anthropic has uncovered the global optima of reasoning prompts.
verdverm 16d ago
I don't really want the models I use learning from Claude at this point. Open weight models of similar scale are available now too, so I expect this "distillation"/"stealing" chatter to wind down.
- hyperpape 16d ago
  Maybe, but that's sort of begging the question that those open weight models aren't significantly trained using "distillation"[0] [0] not technically distillation. https://thomasdullien.github.io/posts/2026-06-15-rl-economics-morally-charged-terms-and-distillation/ https://thomasdullien.github.io/posts/2026-06-15-rl-economic...

trueno 16d ago

[dead]

sippeangelo 16d ago

These draconian "Preserved Thinking" measures they're taking are going to be an absolute pain in the ass. This alone is enough for me to move our API use off their platform entirely. It's a HUGE breaking change that they're trying to dampen by having it not affecting current customers until "in the future", see: https://platform.claude.com/docs/en/build-with-claude/preserved-thinking https://platform.claude.com/docs/en/build-with-claude/preser... You're no longer allowed to edit the context anywhere! The whole context is to become append-only, says Anthropic. No more editing the system prompt as the conversation progresses, no more dynamic loading of custom tool calling formats. Everything has to go through their built-in tools API and you aren't allowed to mess with anything in the context if it has any thinking blocks following it. This is the most intrusive "model DRM" we've seen so far!

solenoid0937 15d ago
You think moving will help you? OpenAI is going to do exactly the same thing soon. Unless you're moving off the frontier entirely, that is.
albert_e 15d ago
How do they still purport to champion alignment and explainablity of AI if the reasoning traces are going to be hidden.
epolanski 16d ago
Interesting, I liked to experiment with a second model "simplifying" and summarizing the previous messages and continue. Needless to say, it improved output on following messages by whatever metric I cared for. Not sure why would they prevent it. I give you a chain of messages, what do you care for what the origin is?
l1n 16d ago
> No more editing the system prompt as the conversation progresses, no more dynamic loading of custom tool calling formats. Hm, aiui you can support both of these via mid-conversation system turns https://platform.claude.com/docs/en/build-with-claude/mid-conversation-system-messages https://platform.claude.com/docs/en/build-with-claude/mid-co... - and in general you'd want to to preserve the cache and recency of the instruction anyways rather than frankensteining an off-distribution transcript. Not sure though.