A fresh wave of AI-safety warnings from industry insiders and policymakers is intensifying pressure for tighter oversight of advanced systems. In the U.K., a parliamentary session organized by Control AI, a lobby group funded by Skype co-founder Jaan Tallinn, featured former Defense Secretary Des Browne and AI scientist Stuart Russell comparing potential superintelligence risks to nuclear catastrophes. Lawmakers weighed a bill to prohibit the creation of artificial superintelligence and discussed pre-release testing, as officials questioned Anthropic’s decision not to submit its latest model, Mythos 5.1, to the U.K.’s AI Safety Institute.
The corporate debate spilled into public view after Evan Hubinger, Anthropic’s alignment science lead, said there is more than a 10% chance advanced AI could kill all humans within a decade and acknowledged the company lacks a definitive plan to ensure alignment. His comments followed the resignation of Anthropic researcher Jacob Coxon, who accused leading labs of “gambling with our lives,” even as he expressed hope for coordination among U.S. firms. OpenAI’s chief scientist, Jakub Pachocki, called for international cooperation on future AI development.
Across the Atlantic, Sen. Bernie Sanders renewed his push for Congressional regulation, citing broad public support. While proponents highlighted existential risks, other experts urged perspective. Academics Andrew Rogoyski and David Barber warned against stifling useful innovation and advocated pragmatic controls and software hardening, and Oxford’s Sandra Wachter argued that job losses, misinformation and environmental costs are more immediate threats than “Terminator” scenarios. Commentator Gary Marcus said outcomes hinge on whether any future superintelligence can be aligned, criticizing hype and limited prudence from major labs. The divide underscores mounting calls for global standards, model testing regimes and enforceable guardrails as frontier systems advance.
Related articles:
Artificial Intelligence Risk Management Framework (AI RMF 1.0)
Introducing the Claude 3 model family
AI Safety Institute (UK) — overview and publications
Pause Giant AI Experiments: An open letter
