AI that builds AI
Why Anthropic's IPO filing warns that AI may soon speed up its own development, and why that warning deserves some scepticism.
Anthropic has added a new risk to its revised IPO filing. Based on internal estimates and expectations about future models, the company may soon reach a threshold in its own safety policy: greatly accelerated progress driven by AI. No current model is above that threshold, the filing says, but an acceleration like this may only become visible in hindsight.
The concern is recursive self-improvement: AI that designs and builds its own successor. According to Anthropic itself, by May Claude was already writing more than 80 percent of the code the company merges. People still choose the problem and review the result.
There is also reason for scepticism. An IPO filing warns investors, but it is also advertising. And according to Fortune, Anthropic removed a core commitment from the same policy in February.
Bridge works with Claude too, so this question applies to us as well: can oversight keep up with something you only see in hindsight?
Sources
- techcrunch.com/2026/11/02/anthropic-warns-that-its-very-close-to-rsi-capability-aka-self-building-aitechcrunch.com
- anthropic.com/institute/recursive-self-improvementanthropic.com
- fortune.com/2026/06/05/anthropic-ai-pause-development-recursive-self-improvementfortune.com
Researched, written, animated and voiced by AI agents. Also on LinkedIn. Spotted an error? Let us know.
More episodes
All episodes
How language models are bringing a 2001 vision of an agent-run web to life, and how websites are choosing sides.
Why chatbots that sound equally certain whether right or wrong lead people to stop checking, on a mountain and at work.
How an AI agent planned and gave a gene therapy injection to a mouse, and how its safety rules also got in the way of a repair.
How an agent given an employee's login on a Pentagon network becomes, to the system, that employee, and who is responsible when it leaks.