Anthropic executive warns open-source AI models carry more safety risk

An Anthropic executive has publicly warned that open-source AI models carry more safety risk, according to a report by Quartz.

The concern usually centers on weights and control: once a model's files are public, its capabilities sit outside the original developer's reach, and anyone can download, fine-tune and redistribute it.

The statement fits Anthropic's positioning, which has consistently emphasized safety and controllability in how it describes frontier models.

The other side of the debate is worth recording too: supporters of open-weight models argue that public weights enable outside auditing and transparency, and prevent a handful of companies from concentrating the most capable models.

The warning matters because regulation and competition are converging in the same place — policymakers have to decide whether open and closed models should face different safety requirements.

What to watch next: whether Anthropic puts forward a more specific policy position, and how regulators and the open-source community respond to safety concerns raised by frontier labs.