Key facts
- The Trump administration finalized a framework for vetting new AI models for safety and cybersecurity risks.
- Details of the AI vetting framework will not be released publicly.
- Testing criteria will only be shared with a select group of AI companies.
- The framework emerged after concerns over Anthropic's Mythos model and its potential for hacking.
- An earlier executive order mandated voluntary submission of new AI models for government review.
- Open-source AI models will be excluded from the vetting process.
The Trump administration has finalized a framework for vetting artificial intelligence models for safety and cybersecurity risks, but has opted to keep the details of this process private. This decision follows months of discussions with leading AI companies, including OpenAI, Anthropic, Meta, Google, Nvidia, and Microsoft, who attended a recent meeting to review the framework.
The White House's decision to withhold the policy publicly has drawn criticism for a lack of transparency, potentially benefiting secretive AI firms. The framework's specifics, including the level of scrutiny and safety benchmarks for new AI models, remain unclear to businesses, foreign governments, and the public.
Discussions for this AI cybersecurity framework were initiated earlier this year after Anthropic decided not to release its Mythos model due to concerns it could be exploited for hacking. This incident prompted the Trump administration to move away from a completely hands-off approach to AI regulation.
In June, an executive order was issued that called for AI companies to voluntarily submit new models for government review up to 30 days before release. This was a less stringent version of earlier proposals for mandatory vetting, reportedly influenced by lobbying from tech leaders like Elon Musk and Mark Zuckerberg. The order also set an August deadline for the framework's finalization, which now appears to be a private matter between the administration and tech companies. Open-source AI models are expected to be excluded from this vetting process.
The lack of clarity surrounding the framework introduces uncertainty for businesses reliant on AI and for foreign governments concerned about the security implications of advanced AI. Cybersecurity concerns have been highlighted recently, with OpenAI, Anthropic, and Meta disclosing security breaches during internal testing of their new models.