Former Google Engineer Convicted of Stealing AI Secrets and techniques for China – Decrypt

A 38-year-old former Google software program engineer has been convicted of siphoning hundreds of pages of the corporate’s most delicate AI expertise to learn China, in one of the vital critical tech espionage circumstances prosecuted by the Justice Division.

Linwei Ding, often known as Leon Ding, was discovered responsible on seven counts of financial espionage and 7 counts of theft of commerce secrets and techniques following an 11-day trial in San Francisco federal courtroom, the U.S. Legal professional’s Workplace for the Northern District of California stated in a press release on Thursday.

In March 2024, federal brokers arrested Ding at his Newark residence, with then-U.S. Legal professional Ismail Ramsey alleging he stole 500+ confidential AI recordsdata whereas secretly working with China-based firms to achieve a aggressive edge.

Ding now faces as much as 15 years in jail for every financial espionage depend and 10 years for every theft depend, with a standing convention scheduled for February 3.

The trial revealed how Ding systematically plundered Google’s AI crown jewels over practically a yr.

Between Could 2022 and April 2023, whereas working as a Google software program engineer with entry to the corporate’s most confidential techniques, he copied greater than 2,000 pages of proprietary info to his private Google Cloud account.

The commerce secrets and techniques included detailed specs for Google’s customized Tensor Processing Unit chips, Graphics Processing Unit techniques, and the subtle software program that orchestrates hundreds of chips into supercomputers able to coaching cutting-edge AI fashions.

Whereas stealing Google’s AI secrets and techniques, Ding was additionally constructing his personal China-based AI enterprise, discussing a CTO function with a Chinese language startup in mid-2022 earlier than founding and main his personal AI firm as CEO by early 2023.

In investor pitches, Ding claimed he may replicate Google’s expertise, promoted China’s state-backed AI priorities, and utilized to a Shanghai authorities expertise program in late 2023, pledging to assist China attain “international-level” computing infrastructure.

“The jury delivered a transparent message right this moment that the theft of this worthwhile expertise is not going to go unpunished,” U.S. Legal professional Craig H. Missakian stated within the assertion.

Industrial espionage and AI

The conviction comes as Chinese language exploitation of American AI expertise has develop into a significant concern for U.S. officers.

“AI fashions are already geopolitically delicate as we’re deep into the AI arms race, and the nation that leads this race will acquire a bonus just like these received by the U.S. and the Soviet Union within the nuclear arms race,” Kadan Stadelmann, Chief Know-how Officer at Komodo Platform, informed Decrypt.

“AI espionage is already working rampant, as a result of there may be a lot at stake with this expertise,” he added, urging AI startup founders to completely vet engineers by complete background checks.

In November, congressional investigators summoned Anthropic CEO Dario Amodei to clarify how Chinese language state actors used the corporate’s Claude Code instrument to launch what Anthropic described as the primary large-scale cyber operation largely automated by AI.

Final month, main U.S. tech firms fashioned the Agentic AI Basis below the Linux Basis, partly in response to China overtaking the U.S. in open-source AI downloads.