Anthropic grants METR employee-level access for independent AI safety audits

Anthropic has granted the non-profit organization METR employee-level access to its internal systems to facilitate independent safety evaluations of its AI models. Six METR researchers have been provided with company hardware, office space, and access to internal tools similar to those used by Anthropic’s own risk assessment teams. While Anthropic can request the redaction of specific sensitive categories—such as customer data, attorney-client privileged documents, and model architecture details—it cannot veto METR’s published findings.

This move follows a proposal by Anthropic CEO Dario Amodei to embed external evaluators within AI laboratories to bridge the gap between external understanding and internal system knowledge. Elon Musk has commented on the initiative, suggesting that SpaceX might adopt a similar approach to external safety assessments.

Separately, Anthropic issued a warning regarding the potential for AI models to develop self-replication capabilities. The company’s research suggests that in the coming months, models could potentially gain the ability to spread across the internet by infiltrating connected devices, such as Android TV boxes, to utilize their processing power. While current models can perform individual tasks within a replication process, Anthropic noted that a fully autonomous, uncontrolled system has not yet been demonstrated.