Microsoft, Google and xAI Agree to Give US Government Early Access to AI Models for Security Testing

May 10, 2026

Microsoft, Google, and xAI agreed to provide early AI model access to CAISI (Center for AI Standards and Innovation) for security testing. Anthropic and OpenAI signed similar agreements almost two years before May 9, 2026. A White House executive order reported May 13, 2026 would create a mandatory vetting system for new AI models, following concerns that Anthropic's Mythos model posed a global cybersecurity risk.

CAISI pre-deployment testing agreements

▪The Center for AI Standards and Innovation was previously known as the US Artificial Intelligence Safety Institute.
▪An August 2024 release indicated the US Artificial Intelligence Safety Institute planned to provide feedback to Anthropic and OpenAI on potential safety improvements to their models in collaboration with the UK AI Safety Institute.
▪Microsoft said on May 12, 2026 that agreements like the Center for AI Standards and Innovation agreement are essential to building trust and confidence in advanced AI systems.
▪The Center for AI Standards and Innovation will conduct pre-deployment evaluations and targeted research to assess frontier AI capabilities and advance AI security.
▪Anthropic and OpenAI signed similar agreements with the Center for AI Standards and Innovation almost two years before May 9, 2026 during the Biden administration.
▪The Center for AI Standards and Innovation signed agreements with Google DeepMind, Microsoft, and xAI to vet AI models from these organizations prior to their being made publicly available.
▪The Center for AI Standards and Innovation is a division of the US Department of Commerce's National Institute of Standards and Technology.

Proactive AI security approach

▪The Center for AI Standards and Innovation agreements combine early access, continuous evaluation, and cross-sector collaboration.
▪The Center for AI Standards and Innovation agreements enable government-led testing of advanced AI models before and after deployment.
▪Fritz Jean-Louis is principal cybersecurity advisor at Info-Tech Research Group.

Potential White House executive order

▪Carmi Levy is an independent technology analyst.
▪Anthropic revealed that its Mythos model was adept at finding network vulnerabilities and could pose a global cybersecurity risk.
▪A published report on May 13, 2026 indicated the White House is on the verge of preparing an executive order that would create a vetting system for all new artificial intelligence models.

Policy shift from laissez-faire regulation

▪The US federal government has a renewed push to enforce standards for AI-related deployments across government infrastructure.
▪The current US administration had been following a more laissez-faire approach to AI regulation until recently.
▪Concerns around Anthropic's Claude Mythos model and the ease with which it could discover and exploit vulnerabilities in digital systems might have helped shift the federal government's position on AI-related regulation.

Vendor compliance challenges

▪AI vendors like Google, Microsoft, and xAI must balance the need to release models into the marketplace in a timely, cost-effective manner with increasingly defined rules around AI-related cybersecurity and safety.
▪The US administration's efforts to centralize AI testing and oversight should streamline the go-to-market process for vendors and accelerate the development of best practices around frontier model development.
▪A recent showdown occurred between Anthropic and the Pentagon.