US National Security Experts Warn AI Giants Aren’t Doing Enough to Protect Their Secrets

Google, in public comments to the NTIA ahead of its report, stated that it anticipates an increase in attempts to disrupt, degrade, deceive, and steal models. The company emphasized that it has a dedicated “security, safety, and reliability organization” comprised of expert engineers and researchers to safeguard its secrets. Additionally, Google mentioned that it is working on a framework that would involve an expert committee to regulate access to models and their weights.

Similarly, OpenAI also expressed the need for both open and closed models in its comments to the NTIA. The organization, known for developing models like GPT-4 and services like ChatGPT, recently formed a security committee and published details on its blog about the security measures it employs in training models. OpenAI hopes that its transparency will encourage other labs to adopt similar protective measures.

During a discussion at Stanford, RAND CEO Jason Matheny echoed concerns about security vulnerabilities, particularly regarding China’s access to powerful computer chips. Matheny suggested that US export controls have restricted China’s ability to develop its own models, leading to an increased likelihood of AI software theft. He emphasized the importance of investing in national security measures to combat this threat.

China’s embassy in Washington, DC, has not responded to WIRED’s request for comment on theft accusations, dismissing such claims as baseless smears by Western officials in the past.

Google reported tipping off law enforcement about a case involving the alleged theft of AI chip secrets for China. While the company maintains strict safeguards against data theft, court documents reveal that it took time to apprehend Linwei Ding, a Chinese national accused of the theft.

Ding, also known as Leon, was employed by Google in 2019 to work on software for its data centers. Prosecutors allege that he copied over 500 files containing confidential information to his personal Google account over a year, evading detection by using tactics like converting files to PDFs and uploading them elsewhere. The US claims that Ding was in contact with the CEO of a Chinese AI startup and had plans to establish his own AI company in China. If convicted, he could face up to 10 years in prison.