- Anthropic says its AI (Claude) now writes >80% of merged code and engineers ship roughly 8× more code per quarter, warning this trajectory could lead to "recursive self-improvement" — AI improving itself without much human oversight.
- The company is calling for an industrywide pause on frontier AI development, but only if multiple top labs agree; the plea arrives as Anthropic nears a $1T valuation and files for an IPO, prompting skepticism about timing and motives.
- Critics note Anthropic recently relaxed a key safety pledge and accuse it of using safety rhetoric competitively; Anthropic counters that the window for public deliberation on AI risks is closing and plans to convene outside stakeholders.
The news, distilled into what matters.
Get the full experience in the app — topics, comments, and audio summaries.