Anthropic prepares report on AI progress and safety practices

Anthropic is getting ready to show its homework. Sam McCandlish said the company will soon publish an analysis of its AI progress and where things stand with its Responsible Scaling Policy.

The timing matters. The disclosure comes as an outside evaluator says Anthropic’s models are improving faster than before, and as the company prepares for a potential IPO.

What Anthropic plans to share

McCandlish made the announcement on November 5, 2026. The report will cover two things: how quickly Anthropic’s AI development is moving, and how the company is tracking against its own safety framework.

The analysis is expected in the coming weeks. Anthropic has not put a firm date on it.

The backdrop is an independent evaluation from METR, an outside group that assesses AI capabilities. METR found that Anthropic’s model capabilities are improving at roughly 1.9 times the rate seen between March 2024 and October 2025.

The research attributes that acceleration primarily to AI helping with research and development, not to AI running the show on its own.

The numbers behind the acceleration

Internal adoption tells part of the story. The value of model usage inside Anthropic grew more than 250-fold between January and September 2026.

The spending is just as striking. Anthropic’s S-1 filing disclosed research and development costs of $2.3 billion in 2024 and $6.3 billion in 2025.

Then came 2026. In the first nine months alone, R&D spending reached $19.2 billion.

Across 2024 through 2026, that adds up to approximately $28 billion. For context, the nine-month 2026 figure is about three times the full-year 2025 total, and more than eight times what the company spent in all of 2024.

A quick primer on the Responsible Scaling Policy

The Responsible Scaling Policy, or RSP, is Anthropic’s internal rulebook for handling risk as its models get more powerful. The company first introduced it in September 2023.

The basic idea works like a set of tripwires. As a model crosses certain capability thresholds, the policy is supposed to trigger stronger safeguards before the company keeps scaling. The goal is to guard against potentially catastrophic outcomes.

Anthropic updated the framework to version 3.0 in February 2026. The revisions emphasize risk assessments, safety protocols, and updated measures for monitoring and controlling AI development.

Anthropic was founded by former OpenAI employees, and safety commitments have long been central to how it presents itself. The RSP is the most concrete expression of that pitch.

Why the IPO makes this report more interesting

Anthropic is preparing for a potential IPO targeted for November 2026. If it happens, it could rank among the largest tech listings on record.

That changes the audience for any safety disclosure. Until now, Anthropic’s RSP updates spoke mainly to researchers, policymakers, and the AI safety community.

The S-1 has already shown how much capital the company is consuming. A report confirming that capabilities are accelerating could help justify that spending, since it implies the money is producing results.

What to watch

The first thing to watch is how Anthropic frames the METR findings. The company could treat the 1.9x acceleration as proof that its research engine is working, as a safety signal requiring tighter controls, or as both.

Second, watch for any disclosure on whether recent models have hit RSP thresholds.

Third, watch the language around autonomy. The research points to AI-assisted research rather than fully autonomous AI as the driver of faster progress.