Anthropic refuses to submit Mythos 5.1 to UK AISI for testing, raising UK concerns over US AI labs' alignment with US protectionism
anthropic
| Source: Techmeme | Original article
Anthropic refused to submit its Mythos 5.1 model to the UK’s AI safety institute for pre‑release testing, sparking British concerns that US AI firms are adopting protectionist stances.
Anthropic has refused to give Britain’s AI Security Institute (AISI) access to its newest frontier model, Claude Mythos 5.1, for pre‑release safety testing. According to the Financial Times, the company limited the testing window to a handful of vetted organisations in the United States, making AISI the first major safety body to be excluded from an Anthropic rollout.
The move has sparked alarm in Whitehall. British officials see the denial as a possible sign that leading U.S. AI firms are aligning with a protectionist stance emerging from Washington, potentially undermining the collaborative safety framework the UK has been building around AISI. If regulators cannot examine cutting‑edge models before they reach the market, the risk of unforeseen harms rises and cross‑border coordination on AI risk mitigation could fray.
The episode adds to a string of recent Anthropic headlines that have highlighted internal safety concerns, from a researcher’s public resignation over “racing” pressures to a class‑action suit over subscription‑tier advertising, which we covered on 9 September 2026. Together, these stories suggest growing tension between rapid product deployment and the industry’s own safety commitments.
What to watch next: the UK government is likely to seek clarification from Anthropic and may consider formal diplomatic or regulatory steps to ensure future access. Industry observers will be watching whether other U.S. labs adopt similar exclusionary practices, and whether the European Union or UK introduces new requirements for foreign AI developers to submit models to accredited safety bodies. The outcome could shape the balance between open innovation and coordinated risk oversight in the global AI race.
Sources
Back to AIPULSEN