On 4 June, Anthropic published an article titled When AI Builds Itself, floating a worldwide "temporary pause" on frontier AI development and committing to convene policymakers to discuss the risks of advanced systems.
The pitch arrives alongside a striking metric: Claude now writes more than 80% of the code that gets merged into its own codebase. Anthropic is framing recursive self-improvement as a near-term governance problem rather than a distant scenario.
What the article said:
Over 80% of code merged into Anthropic's codebase in May was authored by Claude.
Engineers now ship roughly 8x as much code per quarter as in 2021-2025.
In April, Claude shipped 800+ fixes a human would have needed four years for.
On open-ended tasks, Claude's success rate hit 76% in May, up 50 points in six months.
It proposes a coordinated global "brake pedal" on frontier AI.
Introducing Absolutely Agentic Premium
Our premium offering is open for early birds this week before we officially launch. Subscribers will get 3x products for one low price - $15.99 (about £12 in the UK) per month.
Our daily Agentic Intelligence newsletter curates all the best stories from the top AI newsletters, development labs, news websites and social media into one email.
Ad-free podcast - premium ad free RSS feed that you can put into your podcast app of choice. We publish about 3 hours of video a month and now its available in audio form.
Subscriber area includes long form guides to the most important AI trends affecting content and marketing, along with automations and prompts that will save you a lot of time.
We’ll also be adding more content over time, so subscribers always get what’s most important.
Subscribers from this email also get the first three months 50% off on a monthly plan.
Why does this matter?
Recursive self improvement is considered the major step to AGI, and according to this data, we’re not very far off.
This is the first time a major frontier lab has publicly endorsed a coordinated pause as a policy lever, placing Anthropic in unusual alignment with the safety campaigners it has often kept at arm's length.
Our take
Calling for a slowdown is not new in AI safety. The 2023 Future of Life Institute open letter, signed by thousands including Elon Musk, asked labs to pause training beyond GPT-4 for six months. Nobody paused. Anthropic's version is more sophisticated, because it tries to solve the verification problem that sank the earlier appeal, but the underlying move is familiar.
It also helps to be clear about what the post does and does not show. The evidence is about throughput, not a step change in raw capability. Claude is writing far more code and handling longer tasks, yet Anthropic admits the human edge in research taste and judgement persists. The generous reading is that the firm is using its IPO moment to put a question on the table the rest of the industry would rather avoid. The cynical one notes the timing.
Another big thing… Trump eyes a slice of OpenAI
Donald Trump said on 6 June that his administration is in early talks about taking an equity stake in OpenAI, framing it as a way for "the American people [to] benefit from the success of AI." It would be an unusual arrangement for a private US company, and it lands in the same week White House AI advisor Sriram Krishnan announced he is leaving to start an institution shaping Trump's AI policy from the outside.
