AI Safety Coordination
Anthropic urges coordinated global plan to pause advanced AI development amid rising risks.
Anthropic called on Thursday for a global, coordinated plan to pause advanced AI development if risks grow too fast for society to manage, according to Reuters. The AI startup, which just filed confidentially for a $1 trillion IPO, warned that AI systems are beginning to improve themselves — and that humans could lose control if development outpaces safety measures.
The proposal marks a sharp shift for Anthropic. The company previously promised to pause AI development on its own if safety fell behind. But in February 2026, it dropped that pledge, arguing that stopping alone while rivals "blaze ahead" would only hand the lead to less cautious actors, Yahoo Finance reported.
Why Anthropic Says a Solo Pause Won't Work
Anthropic's core argument is simple: unilateral caution is dangerous. If one lab stops and others don't, the cautious lab just loses ground. The company updated its Responsible Scaling Policy in February 2026 to Version 3.0, removing the commitment to pause alone. A further update in May 2026 refined thresholds for chemical and biological weapon risks, according to Reuters.
Now the company wants a collective "pause button" — one that all major labs agree to trigger at the same time. Jack Clark, Anthropic's Head of Policy, warned that AI progress is accelerating toward "recursive self-improvement," where AI builds better versions of itself. He says labs need tools to verify that AI systems stay aligned with human intentions, Yahoo Finance reported.
How Fast AI Is Actually Improving Right Now
The urgency behind the proposal is backed by hard numbers. As of May 2026, 80% of code merged into Anthropic's own codebase was written by its Claude AI — up from low single digits in early 2025, according to Yahoo Finance. The length of tasks AI can reliably complete is now doubling every four months, down from every seven months just recently.
Anthropic's newest model, Claude Mythos, produced 181 successful cyber exploits from several hundred attempts in a test. The previous model, Opus 4.6, produced just 2. One analyst at a Stanford forum warned the model could "wreak havoc" on global banking systems, according to Yahoo Finance.
Critics Say This Is a $1 Trillion Competitive Move
Not everyone is buying the safety framing. Some Silicon Valley rivals and White House officials have accused Anthropic of using safety concerns as a "regulatory moat" — a way to slow competitors while locking in its own dominant position. The timing is notable: Anthropic filed confidentially for an IPO just days before this announcement, at a valuation nearing $1 trillion, Reuters reported.
There is also a technical problem with any pause plan. Verifying that a competitor — especially a state actor like China — has actually stopped training a model is nearly impossible without intrusive on-site monitoring of computing hardware. National security hawks warn that even a coordinated pause could be exploited by Beijing, according to Yahoo Finance.
Summits Planned as Trump Signs AI Safety Order
The announcement lands just two days after President Donald Trump signed an Executive Order on AI Safety on June 2, establishing a voluntary 30-day federal review for advanced AI models. Trump has emphasized a "hands-off" approach to keep the U.S. ahead of China, but acknowledged the need for coordination on national security risks, according to Yahoo Finance.
Anthropic plans to convene a series of summits in the coming months. Policymakers, researchers, civil society groups, and other AI companies will be invited. The goal is to define "verifiable" metrics — specific, measurable thresholds — that would trigger a coordinated halt, Reuters reported. Marina Favaro, Head of the Anthropic Institute, said her team will research the monitoring tools needed to make any global pause credible.
Sentiment
NeutralPublishers
5
Articles
5
Reach
5
