The US Department of Commerce has secured pre-release access to AI models from Google DeepMind, Microsoft, and xAI, joining existing agreements with Anthropic and OpenAI. The companies provide test versions with reduced safety guardrails for classified national security testing.
Five major AI developers have now signed agreements with the Center for AI Standards and Innovation, a division of the Department of Commerce, to provide early access to their models for government security evaluation.
The program, which began with Anthropic and OpenAI, has expanded to include Google DeepMind, Microsoft, and xAI. Each company supplies pre-release versions of their AI systems configured with reduced safety guardrails specifically for testing in classified environments.
The initiative addresses two primary concerns: growing cybersecurity risks and competition with China in artificial intelligence development. By testing these models before public release, US officials can identify potential security vulnerabilities and assess risks related to critical infrastructure, national defense, and intelligence operations.
The reduced safety guardrails in test versions allow government researchers to probe how the AI systems respond to adversarial prompts and potential misuse scenarios. These controlled environments help identify weaknesses that could be exploited by bad actors or hostile nations.
The expansion to five labs represents significant industry cooperation with federal oversight. Each company maintains its own safety standards for commercial releases while supporting the government's security testing needs.
This arrangement reflects broader efforts to integrate AI safety considerations into national security planning. As AI systems become increasingly capable and widely deployed, pre-release access allows government agencies to stay ahead of potential risks.
The program operates within classified settings, limiting public disclosure of specific vulnerabilities or test results. The Department of Commerce has not outlined timelines for regular model submissions or details on how findings inform broader AI policy decisions.
The participating companies continue to develop their own independent safety measures for public-facing products while contributing to government security assessments through this partnership.
A new analysis reveals that calculating the real price of cutting-edge AI models requires multiplying token costs by actual usage patterns. The breakdown challenges how developers and companies evaluate model economics.
Museums are deploying AI chatbots to attract visitors and secure funding, but staff members warn that AI-generated inaccuracies and bias could damage these institutions' credibility as trusted sources of knowledge.
Researchers are flagging a critical risk: widespread AI use in high-stakes professions could prevent workers from developing genuine expertise. The concern centers on whether professionals relying heavily on AI tools will miss essential skill-building experiences.
Microsoft CEO Satya Nadella has raised concerns about companies relying on proprietary AI models from major labs, citing potential vulnerabilities similar to Trojan horse threats.