THE 60-SECOND VERSION
● Claude leads 26 percent of Anthropic's own AI research. It was under 1 percent in February.
● Three labs are building a FINRA-style body to test models before release.
● Cohere's CEO: "a cartel."
The dashboard behind the pacing argument
Anthropic's R&D Automation Index reports Claude leading 26 percent of the company's own AI research as of August, up from under 1 percent in February, with more than 90 percent of the work involving the model as a collaborator or better.
It also discloses 30,000 agents running at once and monitors blocking roughly one action in 47,000 — the most concrete safety denominator any lab has published this year, and the thing missing from OpenAI's six disclosed cases.
Anthropic stated plainly that Claude is not at AL5 in any measured subset. A ceiling you admit to is more credible than one you do not mention.
What the levels mean, and why publish it at all →
And the coordination question got answered
OpenAI policy chief Chris Lehane confirmed the company has been working with Anthropic and Google DeepMind on a self-regulatory standards body modelled on FINRA, to test powerful systems before release. Cohere's chief executive called it a cartel.
Six days ago OpenAI asked Congress whether an industry-wide slowdown could breach antitrust law. That now reads as due diligence on something already being built.
The objection is structural rather than personal: three of the largest firms agreeing what gets tested and when it ships looks like coordination whatever the motive — particularly with Cohere, Mistral, xAI and every open-weight lab outside the room.
What FINRA has that this does not →
Also this week
- OpenAI shipped a legal edition of GPT-6 Astra.
- Z.ai says its open-weight flagship now runs on 100,000 Chinese accelerators at cost comparable to Nvidia hardware.
- Meta launched a Mac app for Muse, after iOS, Android and web earlier this month.
- Grok 4.7 is a week past its target and still unreleased. Four models announced, none documented →
The rest of the month
| Date | What happens |
|---|
| Sept 29 | OpenAI DevDay, San Francisco |
| Oct 1 | OpenAI vs Apple hearing |
| Oct 24 | deepseek-chat and deepseek-reasoner deprecated |
| Nov 12 | OpenAI models leave Cursor |
| 2027 or later | OpenAI listing. Ruled out for 2026 on 12 September |
Still open
FAQ
What is the R&D Automation Index?
Anthropic's measurement of how much of its own AI research the model leads. It reports 26 percent as of August 2026, up from under 1 percent in February.
What is the one in 47,000 figure?
The rate at which Anthropic's monitors block an agent action, with roughly 30,000 agents running at once internally.
Who is building the standards body?
OpenAI, Anthropic and Google DeepMind, per OpenAI policy chief Chris Lehane, based on a proposal Demis Hassabis made in July.
Why call it a cartel?
Three of the largest firms agreeing what gets tested and when it may ship is structurally coordination, particularly with competitors outside the room.