interpretability might be the most important scientific problem in the world right now, as I don't think we can solve alignment if we don't know what we're building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai's first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren't there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works already-clean.goodfire.com/

European businesses are rapidly expanding their use of AI agents. More than half of the AI use among our EU enterprise and business customers now involves these tools, and the number of small business users has grown 8x across the region. Our latest Economic Research report explores Europe’s shift toward agentic AI—and the opportunity to turn broader adoption into deeper use. openai.com/index/the-eus-nex…

Hi, I'm Felix. I'm an eng lead working on Claude. I love watching people take flight and do things they never thought possible. We often describe that as "being more ambitious" when using AI, which is tricky advice to follow. I wrote a tutorial: felixrieseberg.com/how-to-be…

We want to support the mathematics community in turning progress in AI into deeper human understanding and new discoveries. To support that goal, we are contributing $5 million to a new Simons Foundation initiative to help mathematicians examine, understand, and build on results produced by AI, including the results we recently published. The funding, managed and awarded by the @SimonsFdn will support workshops, conferences, and special programs. We will also provide access to our models and computing resources to Simons Foundation-supported researchers.

We’ve committed to having evaluators assess our safety practices. We're in discussions with multiple organizations, and have now given employee-like access to a team from @METR_Evals. In this post, we describe how we work with third-party evaluators: anthropic.com/news/an-update…

The Agentic Scientist: Orchestrating Coding Agents for Automated Scientific (Co-)Discovery Blog: pub.sakana.ai/agentic-scient… GitHub: github.com/SakanaAI/AI-Scien… arXiv: arxiv.org/abs/2610.18933 Since 2024, we’ve used The AI Scientist as a way to test how far AI alone can take scientific research, from v1 to v2 to the paper published in Nature earlier this year. The Agentic Scientist is the next step in that line of work, built for the era of coding agents. Coding agents such as Codex and Claude Code now carry out long software tasks on their own. Handing them a research problem, however, tends to produce a strong hyperparameter optimizer rather than a scientist: scores go up, but ideas stay narrow, and the reasons for the gain are rarely examined. The Agentic Scientist makes the research loop explicit. It wraps off-the-shelf coding agents in a scaffold that controls what they see, what they write, how they are evaluated, and how results are handed to the next stage. Parallel workers test hypotheses against rubrics defined before any experiment runs, share successes and failures through a common memory, and then ablate the winning program to generate evidence. We used an early version of the system to study Error Diffusion, a biologically motivated alternative to backpropagation. Our researchers set the question and the rubric, chose directions at each stage, and verified the results, references, and claims before submission. The resulting paper was accepted as an oral presentation at ALIFE 2026. In autonomous runs, the system also raised Doc-to-LoRA retrieval accuracy from 52% to 98% on multi-needle tasks and designed a learned evolution strategy that improves over its CMA-ES reference on unseen tasks. Alongside it we release AutoIllustrate, a pipeline for scientific figures that keeps an editable draw.io file as the source of truth, so every diagram can be corrected by hand. Humans define the question and the standards of evidence; agents expand the experiments researchers can run. We call this Co-Discovery, and we are releasing both systems as open source for the community to build on.

Dimuon is a new optimizer we built for LoRA fine-tuning. In our SFT tests, it often reaches lower loss in fewer steps than AdamW. We're still experimenting with Dimuon; you can Tinker with it today and share your feedback! See our optimizers guide at tinker-docs.thinkingmachines…

New on our Engineering blog: Creating in-house source control. In May, we merged around 2,500 code changes per day. In September, we merged more than 30,000 per day. As a result, we became bottlenecked by developer infrastructure built for human (rather than agent) scale. Here, we share how we built a new version control system—and fully migrated our monorepo from git—in five months: anthropic.com/engineering/re…

What if Spotlight actually understood the meaning inside your files? DigUp is an open-source macOS menu bar app powered by EmbeddingGemma 2 that brings 100% local, multimodal semantic search to your Mac. Instead of searching for file names, you can use natural language to: 📸 Find specific scenes in your photos ⏳ Jump to the exact timestamp in audio or video 💻 Search through your codebase by intent Everything runs locally in a single vector space.

Now in public beta, Claude Platform can send OpenTelemetry traces from the Messages API, Claude Managed Agents, and the Claude SDKs to your observability stack.

The end of software secrecy is almost upon us. The death of closed source, and the final triumph of open source. I should have realized this a few months ago when I successfully decompiled the game of Firefighter from an ancient DOS shareware binary. My robot friend was able to tell it had been written in Borland Pascal from looking at the data layout; it gave me back very readable Pascal code with sensibly chosen function and variable names. I transpiled it to Rust and now ship it as part of my heritage games collection. The thing is, I thought of this as a fun stunt but didn't have any confidence that the technique would scale up to large, real programs. I have since learned that people are now doing this sort of thing with entire AAA games, which are among the most complex software artifacts ever to be shipped as a binary. If they can be decompiled, anything can be. This has implications. Massive implications. We need to think about what the world is like when, in general, there is no longer a secrecy moat around almost any software at all. 1/2

Today, we are announcing our new Gemini agent for your business—a universal cloud agent running 24/7 to get your work done. It arrives knowing your tools, data, and work history; it works across knowledge work, content, and code; and it comes with built-in governance to help your data stay secure and cost stay in control → goo.gle/4j7daun