Former Google Engineer Sentenced to Nearly One Year for Stealing AI Secrets, Ordered to Compensate Google Nearly $190,000 and Pay $2,500 Fine
Former Google software engineer Linwei "Leon" Ding was sentenced to 12 months minus one day in prison and two years of supervised release by a federal court in San Francisco for stealing trade secrets related to AI supercomputers. He is also required to compensate Google nearly $190,000 and pay a $2,500 fine.
Ding joined Google in 2019 and was part of the team responsible for designing and maintaining the company's AI supercomputer data systems. Evidence presented by prosecutors showed that he downloaded over 2,000 pages of confidential information from Google's internal network between 2022 and 2023 and uploaded it to his personal Google Drive account.
The stolen information involved Google's data center and supercomputer technologies used for training AI models, including technical details about servers, networks, and chip systems. Subsequently, Ding pitched a startup project to investors that aimed to build an AI supercomputer based on Google technology and sought related funding.
A jury found Ding guilty of seven counts of trade secret theft and seven counts of economic espionage after an 11-day trial in January this year. In August, the court ruled that prosecutors failed to prove he knowingly or intended to benefit the Chinese government, thus overturning all seven counts of economic espionage; the seven counts of trade secret theft remained valid.
One of the requirements for economic espionage is that the defendant knows or intends to benefit a foreign government, foreign entity, or foreign agent. The court's ruling did not determine that Ding did not steal the documents, but rather that the evidence was insufficient to meet the higher legal requirement of "serving the interests of the Chinese government"; he was still formally convicted of stealing Google's trade secrets.
Ding had booked a one-way ticket to Beijing before his arrest. Court documents also showed that he had applied for a talent program with the Shanghai municipal government, claiming the plan would help China establish computing infrastructure comparable to international standards; these facts could support his business and return plans, but the court found them insufficient to independently prove his actions met the intent required for economic espionage.
The prosecution had requested a five-year sentence, while the defense sought to replace it with three months of home confinement; the final sentence was significantly lower than the statutory maximum of 10 years in prison and a $250,000 fine for a single count of trade secret theft. The nearly $190,000 compensation amount reflects the court's assessment of Google's losses or investigation costs, not the full commercial value of the stolen AI technology.
From a market mechanism perspective, this case will not directly change Google's AI revenue or chip procurement but will lead U.S. tech companies to further invest in internal access controls, download monitoring, data classification, exit reviews, and access isolation for key R&D teams. Beneficiaries include suppliers of enterprise data security, zero-trust access, endpoint leakage prevention, and internal threat monitoring; employees and partners needing cross-border collaboration and access to core model training and data center materials will face stricter permissions, audits, and compliance costs.
Source: Public Information
ABAB AI Insight
The Ding case occurred during a phase when Google was extending its AI capabilities from general search and advertising technology to large-scale training clusters. After joining the relevant team in 2019, he was exposed not to ordinary application code, but to the data centers, servers, networks, and supercomputer systems required for training cutting-edge models. The competitive barriers in the AI industry thus extend beyond model weights: TPU or GPU cluster design, interconnect networks, storage systems, scheduling, cooling, and power supply architectures also determine unit training costs and model iteration speeds. The commercial significance of stealing this information lies in attempting to shorten the engineering learning curve for building large-scale AI infrastructure from scratch.
The capital path indicates that his actions were not merely for personal collection of materials. Court documents show that while holding Google internal documents, he pitched a startup plan to investors for building an AI supercomputer based on Google technology and applied for a talent program with the Shanghai government. Transforming employer technical materials into a financing narrative implies that he attempted to convert corporate R&D assets into personal entrepreneurial capital; however, legally, whether it can be proven that the ultimate beneficiary is a foreign government, foreign entity, or individual commercial company determines the different sentencing and political attributes of the case falling under trade secret theft or economic espionage.
Historically, core litigation in the U.S. tech industry regarding trade secrets often revolves around the boundaries of "talent mobility" and "technology transfer." Waymo once sued Uber and Anthony Levandowski, accusing him of downloading autonomous driving lidar data before leaving, ultimately settling with Uber paying $245 million in equity; Apple has also taken legal action against former employees for taking autonomous driving project documents. The Ding case shares similarities with these cases in that employees utilize their positions to access core R&D information, but the distinction lies in that the AI supercomputer materials simultaneously touch upon Sino-U.S. technological competition, export controls, and government talent programs, thus leading prosecutors to add economic espionage charges.
This represents a restructuring of the industrial chain. The value chain of cutting-edge AI has expanded from "algorithm papers—software products" to "models—training data—chips—cluster networks—data centers—energy—enterprise customers." As computing infrastructure becomes a core production material, internal design documents are no longer ordinary engineering documents but competitive assets that determine capital expenditure efficiency, model iteration cycles, and market entry speeds. Regulatory and corporate security systems will thus classify and isolate more R&D information: companies will gain stronger technical leak prevention capabilities, but the cost will be an increase in friction for cross-team collaboration and global talent mobility.
ABAB News · Cognitive Laws
Talent can flow, but trade secrets cannot be moved with individuals.
Algorithms can be reproduced, but the engineering path is the most valuable.
Entrepreneurship requires capital; stolen technology will only become evidence.