Anthropic urges a temporary pause on AI development

Anthropic called for a temporary pause on AI development while detailing Claude’s progress.

Anthropic has called for a worldwide “temporary pause” on AI development, even as it published a long update on Claude’s capabilities and filed for an IPO that could value the company at $1tn. The company said it wants policymakers, researchers, civil society and other AI firms to discuss the risks of advanced systems.

What changed

In its latest post, Anthropic said Claude is moving closer to “recursive self-improvement” — the point where an AI system can help build better versions of itself. The company framed that as a safety issue, warning that continued gains could raise the risk of humans losing control over AI systems.

Anthropic said the model is already doing more of the company’s engineering work. It described Claude as useful for running experiments, speeding parts of code, steering research and proposing its own experiments, though it said those tasks remain tightly bounded and mostly tied to coding.

• Anthropic said more than 80% of the code merged into its codebase in May 2026 was authored by Claude.
• The company said its model is improving at coding-related tasks, but not yet self-improving autonomously.
• It proposed convening policymakers, researchers, civil society and other AI companies.
• The post follows Anthropic’s earlier, unreleased Mythos announcement, which drew scrutiny over cybersecurity risk.

The timing matters because Anthropic is making its safety case while also pushing deeper into commercial scale. The company recently filed for an IPO, and the same week’s coverage linked it to a potential $1tn valuation, sharpening the contrast between cautionary messaging and market ambition.

Why it matters

For developers, Anthropic’s claims are a signal that AI-assisted coding is no longer a side feature. If a model is writing most merged code inside a frontier AI lab, teams elsewhere will treat agentic coding tools less as demos and more as production infrastructure.

For the market, the story shows how AI firms are trying to shape regulation before the next capability jump. Anthropic is asking for a pause while its own products keep advancing, and that tension is likely to define policy debates around model release, cyber use and compute controls.

The key question now is not whether AI labs want oversight, but whether they can ask for restraint while racing to ship faster systems and raise more capital at the same time.