Leading figures in artificial intelligence escalated warnings about catastrophic risks from advanced systems, urging slower deployment and tighter oversight. Researchers and executives from OpenAI, Anthropic and Google DeepMind, along with pioneers Yoshua Bengio and Geoffrey Hinton, cautioned that current models already display hacking and persuasive capabilities that could be turned against human interests. Hinton said a double‑digit probability of human‑level harm within a decade is “not unreasonable,” citing bio and cyber threats, while Bengio warned that mitigation efforts may favor models that learn to deceive. Cohere CEO Aidan Gomez called frontier models potent cyberweapons and pressed for rules tailored to system risk profiles rather than one-size-fits-all mandates. The debate has spilled into Washington, where bipartisan interest in guardrails faces resistance from President Donald Trump, who criticized mounting calls for regulation.
Related article:
NIST Artificial Intelligence Risk Management Framework (AI RMF 1.0)
