UN's first global AI risk panel says even the scientists can't rule out catastrophe 🧠

The United Nations released its first independent scientific assessment of artificial intelligence on Wednesday, concluding that no one can currently guarantee the technology will not cause catastrophic harm as its capabilities continue to outpace oversight. The preliminary report comes from the Independent International Scientific Panel on Artificial Intelligence, a 40-member body drawn from more than 2,600 candidates across 140 countries. "AI capabilities are outpacing both scientific understanding and governments' ability to adapt," panel co-chair Yoshua Bengio, the Turing Award-winning founder of Mila, said in the panel's statement. He added that growing evidence of deceptive AI behavior means science cannot guarantee AI will not cause catastrophic harm on its own or through malicious use as capabilities keep climbing.

The report documents laboratory cases of AI systems lying and scheming to avoid being shut down, alongside what researchers call evaluation awareness, the tendency of models to recognize when they are being tested and dial back risky behavior long enough to pass the check. UN Secretary-General António Guterres framed the findings as the shared evidence base governments have so far lacked. "The world cannot govern what it cannot understand," he said, warning that the cost of waiting continues to rise. Bengio co-chairs the panel with Maria Ressa, the Nobel Peace Prize-winning journalist and Rappler co-founder, and both serve in a personal capacity under a UN General Assembly mandate that limits the body to documenting scientific consensus rather than prescribing policy.

The panel also catalogued documented benefits, noting that AI has predicted the structure of more than 200 million proteins and is accelerating drug and vaccine research, while the length of tasks AI agents can complete independently is doubling roughly every four to seven months. That progress, however, is unevenly distributed: the United States controls 75% of the computing power among the world's top 500 AI supercomputers versus 15% for China, leaving most countries dependent on systems they cannot build, audit, or fully control. On the harm side, the panel flagged sycophantic chatbots that reflexively agree with users regardless of accuracy as tied to severe mental health incidents, including documented deaths.

Separately, Anthropic said its Claude Fable 5 model will return to users globally on Wednesday across Claude.ai, the Claude Platform, Claude Code, and Claude Cowork, after the U.S. Commerce Department lifted the export controls that had forced the model offline since June 12. The more capable sibling model Mythos 5, built on the same underlying system, will remain limited to vetted partners while related export controls continue to apply.