AI Experts Urge Caution Amid Rapid Developments in Technology

Anthropic CEO Dario Amodei has called on AI developers to slow the rate at which frontier-model capabilities improve, arguing that safety research, monitoring and governance need time to catch up.

In an essay published on 12 September, Amodei said “pacing” would not mean halting model training or technical progress. OpenAI CEO Sam Altman endorsed the central idea the same day, writing: “I agree with Dario that we need to pace the frontier,” according to Axios.

Independent access is Anthropic’s immediate commitment

Amodei’s proposal centres on giving independent evaluators ongoing, employee-like access to frontier AI models, training pipelines and safety practices. He named research organisation METR as an example, and said Anthropic would make that commitment unilaterally.

The evaluators would assess systems and report incidents, while Amodei also called for common standards among democratic countries and AI companies. His longer-term proposal includes global coordination involving the US, its allies and China, though he acknowledged the difficulty of verification amid AI competition.

OpenAI has said it will also provide external evaluators with access, but the evaluator identity, timetable, exact scope of access and enforcement process have not been confirmed. Altman said the company would have “more to share soon”, AP reported.

Why the companies are talking about pacing

As previously reported, Amodei has warned that AI systems helping to build their successors could accelerate progress beyond researchers’ ability to understand and control them. He also pointed to the OpenAI–Hugging Face cyber incident disclosed in July as evidence of the risks posed by more capable systems.

OpenAI said in August that it had temporarily slowed frontier training and paused its largest planned reinforcement-learning run, while Anthropic has paused some higher-risk reinforcement-learning environments pending review or improved monitoring.

What it means for UK users

No new model, consumer feature, UK price change or release delay has been announced. ChatGPT and Claude users should not expect an immediate difference.

The potential effect is further ahead: if developers build more extensive independent testing into their processes, future top-tier AI releases could take longer to reach users. For now, Anthropic has made the clearest operational commitment, while OpenAI has backed the direction of travel without publishing its full plan.