AI testing center to look for the good and bad in advanced systems for NYS, Hochul says

New York State will conduct its own testing on the safety of advanced artificial intelligence systems, Gov. Kathy Hochul said Tuesday.

"While many AI developers say they’re moving ahead responsibly, the stakes are just too high to take them at their word," Hochul said at a Manhattan news conference announcing the launch of the SUNY Binghamton-based New York Center for AI Responsibility and Research. Hochul said the center would be the first of its kind.

State officials aim to recruit about 80 experts from academic and research institutions to test what Hochul’s office said in a news release were "foundational limits and mathematical theory, formal methods and verification, agentic and complex systems, hardware security and privacy, interpretability and alignment, and scientific AI and biosecurity."

Among its first tasks will be coordinating with the state's Office of Digital Innovation, Governance, Integrity, and Trust to review AI developers' published safety frameworks and provide recommendations on ways to strengthen requirements

The announcement comes amid trillion-dollar valuations for some companies at the AI industry’s forefront and still unanswered questions about risk at the heart of the technology. That includes how human creators can ensure that powerful AI systems act in accordance with human intentions and how vulnerable the systems may be to hijacking by bad actors.

Computer science and cybersecurity experts Greg Morrisett, of Cornell Tech, and Jeannette Wing, of Columbia University, will lead the center as interim co-directors.

Morrisett said center researchers would work on problems like preventing AI agents from escaping "sandboxes," the theoretically contained environments used for testing, and independently evaluate the AI safety measures deployed by developers.

 In an email, Owen Daniels, associate director of analysis at Georgetown University's Center for Security and Emerging Technology, wrote that evaluation would likely be aimed at measuring "risks that AI systems could create, for instance in areas like cyber, biosecurity, or others. Such testing often includes red-teaming practices, data sharing, and model access to understand the conditions under which AI performance degrades or fails." 

In contrast to the government oversight proposed by Hochul, many details about AI and its risks have so far been provided primarily by AI companies. President Donald Trump said in September some of those firms would "self-police" under a voluntary accord that would include internal and external reviews.

Center 'to build tools'

Tom Secunda, chair of Empire AI, the state’s research consortium and computing infrastructure partnership with the state university system and other universities, said the center would be "the first endeavor trying to build tools for people who want to do serious research" and "use science to understand what these models do, how they can be used for good and how they have been and can be used for bad."

AI supercomputers run by Empire AI, Secunda said, are among the most powerful academic computers in the nation and will provide the computing power needed to do that research.

The technology, he added, is "still kind of mystical in some ways ... five to 10 years ago, nobody would have anticipated what these agents can do today."

Under legislation Hochul signed in 2025, large frontier developers are required to register with the state by Jan. 1. They also will be required to publish safety frameworks on their websites, report critical safety incidents within 72 hours and publish model transparency reports, Hochul’s office has said.

 Daniels, of Georgetown, wrote that evaluating frontier models would take "technical expertise to assess the models' capabilities, information sharing with the government from the leading AI labs, and careful risk assessments in different use contexts. The Center appears focused on bringing talent to bear on these issues."

Other states including California, Texas and Utah have passed their own AI laws in recent years, he wrote. "As a new Congress contemplates what a federal AI framework might look like in 2027, state laws are bridging the gap and providing laboratories for policy experimentation." 

With The Associated Press

By Nicholas Spangler

nicholas.spangler@newsday.comspanglernewsday

Nicholas Spangler is a general assignment reporter and has worked at Newsday since 2010.