Meta is positioning its new Muse Wonder models as top-tier on benchmarks including coding and agentic tasks, marking its entry into the frontier lab race.

Smith used the milk and washing machine comparison to argue that voluntary AI safety commitments are insufficient and that binding oversight is needed.

The fabricated tip highlights growing concerns about agentic AI systems taking real-world actions without adequate safeguards or accountability.

The Maia 300 rollout would reduce Microsoft's reliance on Nvidia GPUs and cut inference costs across its products.

The reversal restores the UK AI Security Institute's pre-release access to models from Anthropic, Google DeepMind, and OpenAI.

The departures add to scrutiny of how frontier labs balance internal safety dissent against commercial priorities.

More than 40 publishers are pursuing copyright claims over the use of their content to train generative AI systems.

The listing would be a landmark test of investor appetite for Chinese AI companies facing both legal and geopolitical headwinds.

Karpathy points to a widening gap between the few thousand people using internal frontier systems and the vast majority who have barely touched LLMs.

Security officials and Bill Gates separately cautioned that AI without safeguards could accelerate offensive cyber operations.

The shift would place federal AI evaluation staff closer to the White House AI task force weighing safety and competitiveness.

The column warns that the scale of capital committed to AI has made a correction systemically significant.

The online reinforcement learning system trains coding agents on unmodified environments over extreme-long horizons.

Lawmakers are weighing who bears responsibility when autonomous agents cause harm, a gap courts and regulators have yet to settle.

The arrangement gives British evaluators access to frontier models before public release, filling a gap left by US policy shifts.

A wave of commentary argues existing liability frameworks were not written for autonomous systems that act on their own.

The expansion deepens OpenAI's European footprint as it builds out customer-facing deployment teams.

Tracking shows roughly $80M spent on nearly 170 AI-generated political ads this year, raising questions about disclosure rules.