Overview
- On June 10 Anthropic published two frameworks and an essay that call for mandatory independent audits of 'frontier' models and expanded access to its newest models, Claude Mythos 5 and Claude Fable 5.
- The safety plan would require third‑party testing across four risk areas—cybersecurity, biological threats, loss of control, and automated research—and give governments the legal power to block or reverse deployments judged unsafe.
- Anthropic's own Mythos testing found thousands of high‑severity software vulnerabilities that could be exploited by models, a result the company says shows acute dual‑use cyber risks to critical systems like hospitals.
- The company also proposed economic measures to address job disruption, including better data collection, wage insurance, retention tax incentives, workforce training, and options such as universal capital accounts or long‑term income support.
- Responses were mixed: some regulators and safety researchers welcomed stronger rules, critics including OpenAI's Sam Altman warned the approach could be used to concentrate power, and experts flagged verification, auditor design, enforcement, and international coordination as major hurdles.