Key facts
- Senator Mark Kelly supports the Trump administration's directive for Anthropic to suspend access to its Fable 5 and Mythos 5 AI models.
- The administration cited security concerns and a potential method for bypassing the models.
- Anthropic complied with the directive but stated it was unnecessary and did not align with transparency principles.
- Defense Secretary Pete Hegseth previously prohibited Anthropic's AI chatbot within the Pentagon.
- President Trump signed an executive order for voluntary government oversight of AI models before public release.
Senator Mark Kelly has aligned with the Trump administration's directive for AI firm Anthropic to suspend access to its Fable 5 and Mythos 5 models, citing security concerns. Kelly, a member of the Senate Armed Forces Committee, stated that AI companies must be cautious when releasing new models, and that Anthropic's willingness to work with the government on this issue is a positive sign.
The directive followed reports that a method for bypassing the Fable 5 model, known as 'jailbreaking,' had been discovered. Anthropic complied with the order, disabling access to the specified models, but argued that the directive was unnecessary and did not adhere to the transparent, fair, and technically grounded principles outlined in President Trump's earlier executive order on AI oversight.
This latest clash comes after Defense Secretary Pete Hegseth previously labeled Anthropic as a supply chain risk and prohibited its Claude AI chatbot within the Pentagon. Anthropic has sued the administration over that designation. David Sacks, a former special adviser to Trump on AI, criticized Anthropic's response to the latest directive, suggesting it was at odds with the company's stated commitment to AI safety.