AI Newsletter

Anthropic officially confirms that Claude already writes over 80% of its own source code – and the new model Claude Mythos was classified by METR as potentially too powerful for public release. This marks a qualitative leap in recursive self-improvement that goes far beyond previous coding assistants. The finding is not only technically explosive, but has triggered direct regulatory consequences and reignited the entire safety debate.

Jeff Bezos' AI startup Prometheus has completed a new funding round that values the company at $41 billion, making Bezos one of the most significant individual players in the AI race. The deal signals that capital flowing into the AI industry remains undiminished despite growing skepticism about missing EBIT effects. For the market, this means further concentration of power among a few capital-rich players outside traditional lab structures.

Google DeepMind has published internal concerns about what happens when millions of autonomous AI agents interact simultaneously – a scenario that is becoming increasingly realistic with the rapid rollout of agentic-based enterprise systems starting in 2026. The warning is unusually open for a leading lab and points to systemic risks that are covered neither by single-model security tests nor by existing regulation. For companies currently switching to agentic workflows, this significantly increases pressure on governance frameworks.

Cognition, maker of AI software engineer Devin, has raised over $1 billion at a valuation of $26 billion and reports more than tenfold enterprise growth since the beginning of 2026. This is the strongest data evidence to date that AI-powered software development has reached the tipping point from experiment to core infrastructure in the enterprise segment. The move increases competitive pressure on established dev tool providers and developer labor markets alike.

For the first time, researchers within OpenAI are publicly signaling support for a global AI pause call – whereas such demands previously came primarily from Anthropic employees and external safety researchers. The thread has high organic reach and shows that the safety debate is now capturing leading commercial labs from within. This could accelerate regulatory processes and sustainably shift internal governance culture at top labs.