Emmanuel Ameisen

We've made progress in our quest to understand how Claude and models like it think! The paper has many fun and surprising case studies, that anyone who is interested in LLMs would enjoy. Check out the video below for an example

We evaluated an early version of Claude Mythos Preview for risk assessment during a limited window in March 2026. We estimated a 50%-time-horizon of at least 16hrs (95% CI 8.5hrs to 55hrs) on our task suite, at the upper end of what we can measure without new tasks.

@__lightyear__ The reason the NLA showed these results on opus is that it was trained on transcripts where it ended up needing to infer the user's language. That's not true for the neuronpedia models (paper has more details)

@mlpowered me and a few other people on HN are saying that the examples you showed off with opus don't work with the models on neuronpedia, like the russian 'mom is sleeping in the next room' prediction. the haiku examples work though. model too weak?

Interpreting model activations is important to understand why a model is doing what its doing. Traditionally, we've done this with supervised methods (probing for a specific context), or unsupervised sparse decompositions (dictionary learning). But probing requires you to know what you are looking for, and sparse dictionaries can be overwhelming to interpret. NLAs are exciting because they instead generate natural language explanations, which we can then inspect for a variety of behaviors. For example, they reveal the planning behavior we first observed with circuit tracing last year. They also helped identify bugs in Claude's training pipeline, where some prompts were only partially translated. If you want to play with them, NLAs on open models are available on Neuronpedia! neuronpedia.org/llama3.3-70b-i…

Interpreting language models can feel like stumbling through a dark forest - sometimes you just wish you had a flashlight! In our new post, we introduce HeadVis, our latest flashlight for studying attention heads.

How do LLMs store attributed of entities? And how do they compare different attributes in context? It turns out they mostly store information about a given entity over its own token, which allows for easy lookups. But in addition to the current entity's information, models also store information about the previous entity. That might seem redundant, but it actually enables a model to identify relationships between the current entity and the previous entity in one step!

Do LMs plan without verbalizing their plans? I'll be at ICLR presenting work with @mlpowered using circuit tracing to reveal latent planning—from choosing "a" vs "an" based on a planned-for word, to rhyming poetry—and how these abilities grow with scale: openreview.net/forum?id=H0B7p…

Made this 30 second video of Claude Design just by pasting in the Claude Design blog post and some tweets from @AnthropicAI employees Kinda speechless.

Anthropic’s Opus 4.7 just seized the #1 spot on the Vals Index with a score of 71.4%, a massive jump from the previous best (67.7%). It also ranks #1 on Vibe Code Bench, Vals Multimodal, Finance Agent, Mortgage Tax, SAGE, SWE-Bench, and Terminal Bench 2.

🧵New Anthropic Fellows research: We studied mechanisms of "introspective awareness" in LLMs. LLMs can sometimes detect steering vectors injected into their residual stream. But is this worthy of being called introspection, or attributable to some uninteresting confound?👇

Introducing Project Glasswing: an urgent initiative to help secure the world’s most critical software. It’s powered by our newest frontier model, Claude Mythos Preview, which can find software vulnerabilities better than all but the most skilled humans. anthropic.com/glasswing

@AndrewLampinen Welcome!

Career update: I joined Anthropic (alignment team) this week — exciting place to be at an exciting time!

We partnered with Mozilla to test Claude's ability to find security vulnerabilities in Firefox. Opus 4.6 found 22 vulnerabilities in just two weeks. Of these, 14 were high-severity, representing a fifth of all high-severity bugs Mozilla remediated in 2025.

@oanaolt Appreciate you Oana!

@mlpowered As I’ve told Benn, thanks for respecting our freedoms and for the work you do. Here to support any way I can.

Proud to work at a place that stands behind its values. 🇺🇸

I used to bite my tongue and hold my breath. Scared to rock the boat and make a mess. I stood for nothing, so I fell for everything. 🎶

AI is not a normal technology, and Anthropic’s mission is to make sure that it serves the long-term benefit of humanity. Doing so requires making tough decisions, and standing up for what we think is right. This is us doing that.

Late last year, we found a precise counting mechanism in Claude. This new work by @ummagumm_a and Nikita Balagansky shows that: - similar mechanisms exist in many models - we can compare their counting performance by seeing how crisp their representations of the count are!

@ummagumm_a @wesg52 @ch402 @thebasepoint @AnthropicAI @neuronpedia @tfrere Very cool work! Did you get a chance to look at the boundary estimation mechanism? It'd be interesting to know if performance diffs are explained by ability to: - estimate line position/width - combine both to know how many chars are left - know the length of the next token

cc: @wesg52, @mlpowered, @ch402, @thebasepoint, @anthropicai, @neuronpedia, @tfrere

1/ 🧵 Reproducing Anthropic’s “counting manifold” result in open-weight LLMs: do they internally track “chars since last \n” to wrap text consistently? huggingface.co/spaces/t-tech/…

Yesterday's OpenAI and Anthropic Terminal-Bench 2.0 results used different harnesses. Run both in Terminus 2 ➡️ ~similar scores (within noise). Harnesses matter! Congrats to both teams on incredible models!

We recently released a paper on Activation Oracles (AOs), a technique for training LLMs to explain their own neural activations in natural language. We piloted a variant of AOs during the Claude Opus 4.6 alignment audit. We thought they were surprisingly useful! 🧵