TECH GIANTS TO ALLOW PRE‑RELEASE GOVERNMENT TESTING OF AI MODELS AS CYBERSECURITY CONCERNS SURGE
THE PRESS CENTER | THE MORSE CODE | TECHNOLOGY
MAY 9 2026 AT 5:31 PM
Google, Microsoft, and xAI have agreed to provide the U.S. government with access to unreleased versions of their most advanced AI systems—a significant shift in how frontier AI will be evaluated before reaching the public. The National Institute of Standards and Technology (NIST) announced the arrangement Tuesday, framing it as a critical step toward preventing AI‑driven cybersecurity threats before they materialize.
The move follows a turbulent month in the AI sector. Anthropic’s new Mythos model—described by the company as substantially ahead of existing systems in cybersecurity capabilities—triggered alarm across government agencies, financial institutions, and critical infrastructure operators. Its emergence accelerated discussions inside the White House about whether the United States needs a formal, government‑run review process for powerful AI models, similar to how pharmaceuticals or advanced weapons systems are evaluated.
A New Federal Role in AI Oversight
Under the new agreements, the Center for AI Standards and Innovation (CAISI)—a unit within the Department of Commerce—will gain early access to cutting‑edge AI models. This includes the ability to:
- Evaluate national security and public‑safety risks before models are released
- Conduct post‑deployment testing to identify vulnerabilities that emerge over time
- Scale up technical research using industry‑provided compute and expertise
CAISI has already completed more than 40 evaluations of existing AI systems, but the agency has long acknowledged that its resources lag far behind those of major tech companies.
Chris Fall, CAISI’s director, emphasized the stakes:
“Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications,” he said, noting that expanded industry cooperation arrives “at a critical moment.”
Mythos and the Tipping Point for AI Security
Anthropic’s Mythos model is widely seen as the catalyst for this new posture. The company has stated that Mythos is so capable in cybersecurity—both defensive and potentially offensive—that it is not comfortable releasing it broadly. Instead, access is restricted to a small group of vetted organizations, and senior U.S. officials have already been briefed on its capabilities.
The episode underscored a growing fear: that AI systems could soon automate or accelerate cyberattacks at a scale that existing defenses cannot match.
OpenAI, facing similar concerns, announced last week that it will provide its most advanced models to all vetted levels of the U.S. government, aiming to stay ahead of AI‑enabled threats rather than react to them.
Why the Government Needs Industry Help
Jessica Ji, a senior research analyst at Georgetown University’s Center for Security and Emerging Technology, noted that CAISI’s partnership with major AI labs is not just beneficial—it is necessary.
She pointed out that the federal government lacks the compute power, specialized personnel, and technical infrastructure required to rigorously test frontier‑scale AI models on its own.
“They simply don’t have the same amount of resources as big tech companies,” she said.
The partnerships effectively give CAISI a window into systems that would otherwise be too large, too complex, or too expensive for the government to evaluate independently.
A Shift in Federal AI Policy
The White House is now consulting experts on whether the U.S. should adopt a formal review process for advanced AI models—something that would mark a clear departure from the previous administration’s hands‑off approach to AI regulation.
A White House spokesperson stressed that no decisions have been finalized:
“Any policy announcement will come directly from the President. Discussion about potential executive orders is speculation.”
Still, the creation of an expert working group—first reported by The New York Times—signals that the administration is preparing for a more assertive regulatory role.
Industry Reactions
Microsoft, which already conducts extensive internal testing of its models, welcomed the collaboration. Chief Responsible AI Officer Natasha Crampton said CAISI brings “technical, scientific and national security expertise” that complements Microsoft’s own evaluations.
Google declined to elaborate on its participation, and xAI did not respond to requests for comment.
SOURCE CREDIT: CNN.COM
