Microsoft, Google and xAI grant US access to AI models for testing
Updated
Updated · Al Jazeera English · May 5
Microsoft, Google and xAI grant US access to AI models for testing
12 articles · Updated · Al Jazeera English · May 5
The Commerce Department's CAISI said the companies will let federal officials evaluate new models before deployment, amid fears Anthropic's Mythos could aid hackers.
CAISI, the government's main AI testing hub, said it has completed more than 40 evaluations, including on unreleased systems, often with safety guardrails reduced to probe national security risks.
The move extends 2024 deals with OpenAI and Anthropic and follows a separate Pentagon pact with seven tech firms to use AI on classified networks.
With AI now able to autonomously hack systems, how can we ensure it remains a tool for defense, not attack?
With AI development costs soaring, is the future of powerful AI controlled by only a handful of corporations?
Anthropic Barred as Pentagon Pushes $200M AI Integration with Google, Microsoft, OpenAI, and Others
Overview
In early 2026, the Pentagon formed strategic partnerships with eight leading AI companies to create an "AI-first fighting force," deploying advanced AI tools through the GenAI.mil platform to over 1.3 million personnel. This effort required vendors to accept a "lawful use" standard allowing unrestricted military applications, including surveillance and autonomous weapons. While most companies agreed, Anthropic rejected this on ethical grounds, leading to its designation as a supply chain risk and a subsequent lawsuit challenging the government’s action. Meanwhile, industry tensions grew as employees and investors pressured firms like Google and OpenAI over ethical concerns. Simultaneously, xAI’s Grok 4 advanced AI capabilities gained traction in defense and enterprise, and OpenAI’s shift to multi-cloud partnerships intensified competition among cloud providers, reshaping the AI landscape amid ongoing debates over military AI governance.