The U.S. government has secured agreements with Google, Microsoft, and Elon Musk’s xAI to evaluate their latest AI models for national security risks before public release. The Center for AI Standards and Innovation (CAISI), under the Department of Commerce, will conduct pre-deployment assessments to identify potential cybersecurity threats and misuse risks.
Core Developments:
- CAISI announced agreements with Google, Microsoft, and xAI on May 5, allowing early access to AI models for security testing.
- The move follows existing partnerships with OpenAI and Anthropic, which were renegotiated under the Trump administration’s AI Action Plan.
Deeper Context:
Broader Oversight Efforts: The White House is considering a new AI working group to formalize oversight procedures, potentially through an executive order. This group would involve tech executives and government officials to vet models before public release.
Anthropic’s Role: Anthropic’s Claude Mythos Preview model, which excels at identifying software vulnerabilities, has drawn scrutiny. The company limited its rollout to select firms under Project Glasswing, a cybersecurity initiative. Anthropic CEO Dario Amodei met with Trump administration officials amid concerns over its capabilities.
Pentagon Dispute: Anthropic is locked in a legal battle with the Pentagon over safeguards for military AI use. Defense Secretary Pete Hegseth labeled the company a “supply chain risk” after it resisted loosening restrictions.
CAISI’s Role: CAISI, formerly the U.S. AI Safety Institute, has completed over 40 evaluations of cutting-edge models, including versions with stripped safety guardrails. The agency aims to assess risks before models reach the public.
Industry Response: Microsoft and xAI did not immediately respond to requests for comment, while Google declined to comment. The Pentagon recently reached agreements with seven AI firms to deploy their tools on classified networks.