AI Safety Demands Concrete Goals, Not Just Slower Timelines
A prominent AI researcher argues safety standards must meet firm benchmarks amid growing industry turmoil over safety practices.
Tensions over artificial intelligence safety have reached a new pitch following the resignation of AI safety researcher Jacob Coxon from Anthropic, one of the industry's leading AI development firms. The departure has reignited a broader debate about whether the technology sector is doing enough — and the right things — to ensure AI systems do not pose unacceptable risks to the public.
Stuart Russell, a leading voice in AI research, contends that genuine safety cannot be achieved simply by slowing the pace of AI development. Instead, he argues, safety requirements must be grounded in specific, measurable outcomes — concrete goals that developers are obligated to meet before advancing their systems further.
Read more UK Unions Praise Labour Team but Brace for Rights Battles →
The controversy has been compounded by a separate and ongoing controversy involving OpenAI and Hugging Face, with weeks of what observers have described as increasingly alarming revelations emerging from that situation. The convergence of these events has pushed the AI safety debate into mainstream business discourse, with outlets describing the moment as an "AI doomsday debate" reaching a boiling point.
The stakes of this debate extend well beyond the technology industry. Policymakers, investors, and the public are increasingly attentive to how AI companies govern themselves internally on safety matters, and whether voluntary commitments are sufficient or whether binding standards are necessary. Russell's position reflects a growing school of thought that process-based assurances — such as pledging to move carefully — are insufficient substitutes for verifiable safety benchmarks.
Continue reading at Business | The Guardian.