Exclusive: Senators interrogate AI firms on China safeguards
Add Axios as your preferred source to
see more of our stories on Google.
Illustration: Brendan Lynch/Axios
Sens. Chuck Grassley (R-Iowa) and Jim Banks (R-Ind.) are demanding answers from major tech and AI companies over concerns that employees with ties to China could access cutting-edge U.S. AI systems.
Why it matters: Lawmakers are focusing on insider access — not just hacking — as a potential vulnerability, putting pressure on companies to demonstrate stronger safeguards.
- "The Chinese Communist Party (CCP) has an extensive track record of conducting espionage on U.S. companies in critical sectors," the senators write.
Zoom in: Identical letters sent to the CEOs of Amazon, Anthropic, OpenAI, Google, Meta, Microsoft, Safe Superintelligence Inc., Thinking Machines Lab and xAI ask for responses to nine questions by May 20.
- One question zeroes in on insider risk: "What is Amazon's current approach to personnel vetting, insider threat detection, and ongoing monitoring for privileged access roles?" the letter to CEO Andy Jassy asks.
- "How many PRC nationals does Amazon employ? How many have direct or indirect privileged access to weights or weight-adjacent assets, and how has that number changed over time?" reads another question.
Between the lines: The senators also indicate they want to work collaboratively with companies to identify and plug potential security gaps.
- "What support or engagement from Congress or the U.S. government would be useful in securing AI technology, trade secrets, and research from the PRC?"
Zoom out: Congress has ramped up oversight, including classified briefings with leading AI firms, as Axios reported Tuesday.
- U.S. officials have warned that China is conducting "industrial-scale" efforts to extract American AI capabilities, underscoring bipartisan concerns about espionage and technological competition.
- Lawmakers have also increasingly engaged directly with companies like OpenAI, Anthropic and Nvidia on how their technologies could be secured against foreign access.
