Anthropic Skips UK Pre-Release Tests for Mythos 5.1, FT Reports
Anthropic did not provide its Mythos 5.1 model to Britain’s AI Security Institute (AISI) before its launch, according to the Financial Times (FT). This is a first for a major AI model. UK officials are concerned that US labs are adopting a protectionist approach.
Key Points:
- Anthropic granted access to comparable US organizations but not AISI.
- The UK institute had previously tested Anthropic models, including an evaluation of its cyber capabilities in April.
- Anthropic released Mythos 5.1 and Fable 5.1 on September 1st, limiting access to approved partners in Project Glasswing.
- ENISA, the EU’s cybersecurity agency, began testing Mythos 5 on Thursday, but still does not have access to the latest version.
- The UK government emphasized AISI’s continued collaboration with industry partners, including Anthropic.
- Policy analysts predict a broader trend of US AI companies prioritizing security as a competitive advantage.
Quote:
"These risks do not stop at national borders and no country can tackle them alone," said a Cabinet Office spokesperson.
In summary, the decision by Anthropic to skip UK pre-release testing raises concerns about international collaboration in AI safety and the potential for protectionist measures.