US Expands Safety Tests for New AI Models

The US government is widening its scrutiny of advanced artificial intelligence, striking new agreements with Google, Microsoft and xAI to safety test emerging models before they spread further into public life.

The move builds on voluntary commitments forged during the Biden administration, and it signals that federal officials want a more structured view into how powerful AI systems behave. Reports indicate the Commerce Department will use the agreements to examine model risks, extending a framework that already drew in leading developers as Washington raced to keep pace with the technology.

The new agreements show Washington is trying to turn broad AI promises into a repeatable safety process.

The timing matters. AI companies keep releasing more capable systems at a breakneck pace, while regulators and lawmakers still struggle to define clear rules for testing, deployment and accountability. By pulling more companies into federal review, the administration appears to be betting that early access and technical evaluation can reveal dangerous flaws before they hit consumers, businesses or critical systems.

The agreements also highlight a basic tension at the center of AI policy: the government wants innovation, but it also wants visibility into tools that could produce serious harm if left unchecked. Sources suggest officials see these partnerships as a practical way to gather evidence while broader legal and regulatory debates continue. For the companies, cooperation may help shape whatever standards come next.

What happens next will matter far beyond Silicon Valley. If these safety tests produce credible benchmarks, they could influence future federal rules and even set expectations for the wider industry. If they fall short, pressure will grow for tougher mandates. Either way, the message is clear: the era of releasing frontier AI with minimal outside review is facing a stronger challenge from Washington.