Product1 distinct publisher3 min readUpdated
Seven economic espionage convictions against a former Google engineer are gone; seven trade secret convictions remain. The evidence gap was allegiance, not exfiltration.
The Product Desk · Product desk

Compiled by The Product DeskSomething wrong?How this is made
A federal judge has thrown out seven economic espionage convictions against former Google engineer Linwei Ding while leaving seven trade secret theft convictions standing, finding that prosecutors proved what he took but not that he was working for China [1]. For anyone who owns insider risk at a company with valuable model or datacenter designs, the ruling is a map of which half of the standard narrative survives contact with a courtroom.
Judge Vince Chhabria found the evidence sufficient to show that Ding took the material, and insufficient to show he intended or knew his conduct would benefit the Chinese government [2]. A jury had convicted him on all 14 counts in January after an 11-day trial [3]. The statutory difference is narrow and total: economic espionage requires a foreign state connection, theft of trade secrets does not [4]. According to the account of the ruling, Ding had been accused of connections to Chinese technology companies rather than to the state itself, and that is the gap the decision turns on [5]. A commercial recipient in China is not the government of China [6].
What was taken is not contested. Prosecutors said Ding removed thousands of pages of confidential material on Google's AI infrastructure, including hardware blueprints and the software platforms the company uses to train large language models in its datacenters [7]. That is the tensor processing unit stack, the part of Google that is genuinely difficult to replicate and the reason the case drew attention [8].
The arithmetic of exposure explains why the ruling matters to Ding and not only to prosecutors. The espionage counts carried up to 15 years each and the theft counts up to 10 [9], so the vacated counts represented a nominal maximum of 105 years against 70 remaining [10], and the decision removes the heavier half of his exposure entirely [11]. Sentencing is set for 1 September, with up to 10 years on each of seven counts, and the practical outcome will turn on how the guidelines treat the value of what was taken and whether the court finds he intended to use it commercially [12].
The procedural lesson is the durable one. Intent to benefit a foreign government is usually proved through communications, payments, or a documented relationship, and here the jury convicted without the judge being satisfied that any of that had been established [13]. Charging the state connection is attractive because it converts a commercial dispute into a national security matter, and the Justice Department has been building that case in public [14]. The White House has accused China of industrial-scale model distillation and committed to sharing intelligence with OpenAI, Anthropic, and Google, an unusual arrangement between a government and a set of private companies [15]. Enforcement has been busiest at the hardware end, where Taiwanese prosecutors have detained an Nvidia employee in a chip-smuggling investigation and Super Micro's co-founder has been charged over server exports to China [16].
None of this changes Google's position. The company detected the exfiltration, reported it, and testified, and the material is gone regardless of which statute applies [17]. Detection and documentation are what the company controlled; the foreign-state element was never in its hands.
Watch whether the Justice Department appeals. Prosecutors can appeal a post-verdict acquittal of this kind, but the standard is demanding, because the judge is finding that no reasonable jury could have reached that conclusion on the evidence presented [18]. Ding's lawyer welcomed the ruling; the department and Google did not immediately comment, and neither has said whether an appeal is planned [19]. Watch, too, whether future indictments against engineers lead with the trade secret counts and treat the state connection as an add-on rather than the frame.
Follow any of these and your For You feed starts watching them — no settings page required.
Ranked by verification strength, evidence, and original report placement.
Prosecutors said Ding removed thousands of pages of confidential material on Google's AI infrastructure, including hardware blueprints and the software platforms the company uses to train large language models in its data centres.
The material was the tensor processing unit stack, the part of Google that is genuinely difficult to replicate and the reason the case drew attention.
A federal judge threw out seven economic espionage convictions against former Google engineer Linwei Ding on the grounds that prosecutors never proved he was working for China; seven trade secret theft convictions stand.
Judge Vince Chhabria found the evidence sufficient to show that Ding took what he took, but insufficient to show that he intended or knew his conduct would benefit the Chinese government.
A jury found Ding guilty on all 14 counts in January, following an 11-day trial.
Economic espionage under US law requires a foreign state connection; theft of trade secrets requires no such thing.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Specific and internally consistent, but single-source and second-hand
The core facts are concrete and checkable in principle: named judge, count-by-count outcome, 14-count January verdict after an 11-day trial, per-count statutory maxima, and a 1 September sentencing date. Against that, the cluster has one publisher and no direct quotation from the ruling, no docket citation, and no DOJ or Google response, so the court's reasoning reaches the reader entirely through the outlet's paraphrase. Analytical generalisations about how foreign-benefit intent is normally proved are asserted without support.
Real enforcement footprint, doctrinal effect not yet observed
There are concrete real-world events beyond the ruling itself: a detected and reported corporate exfiltration that reached trial with company testimony, a chip-smuggling detention involving an Nvidia employee, charges against Super Micro's co-founder, and a government-to-lab intelligence-sharing commitment. That is a visible enforcement pattern. What is not yet observable is any uptake of the ruling's narrowing effect: no second case charged differently, no appellate treatment, no sentencing outcome, and no DOJ statement of changed charging practice.
Restrained reporting, but the precedent framing outruns one district ruling
The prose itself is careful and repeatedly deflates the geopolitical reading, noting the ruling neither vindicates Ding nor clears Google's security posture. The overstatement is structural: presenting a single post-verdict acquittal, still appealable and with sentencing pending, as narrowing how insider cases get charged assumes a doctrinal ripple no source in the cluster demonstrates. The unsourced generalisation about how foreign-benefit intent is normally proved pushes the same direction.
Institutional framing incentives on all sides, plus outlet distribution interest
The story's sourcing sits inside a field of interested parties: the Justice Department has been publicly arguing that AI trade secret theft is a national security matter and staked that framing on this case, the White House has an interest in the industrial-scale distillation narrative and in binding named labs into intelligence sharing, Google benefits from being seen as the detecting and reporting victim, and the only on-record reaction is from defence counsel welcoming the ruling. The publisher is a commercial tech outlet closing with a newsletter solicitation. None of these incentives is hidden, and the article names the DOJ's awkward timing rather than concealing it.
Facts likely accurate; interpretation weakly corroborated
Confidence in the narrow factual core is fairly high because the details are specific and unlikely to be misreported. Confidence in the story as told is lower: one publisher, no primary document, no institutional response, an unresolved appeal question, and a headline thesis about charging practice that no second source or legal authority in the cluster tests. Sentencing on 1 September is a near-term event that will materially update the picture.
product
Washington's secret AI test is coming for open weights, and release dates go with it2 distinct publishers
invest
Google Ships Flash Instead of Pro While OpenAI Loses Its Two Best Operators1 distinct publisher
security
The nationalization argument is really a vendor-continuity memo1 distinct publisher
product
AMD borrows $4.75bn while sitting on $13bn, and the number matches its Anthropic promise1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
1 article · August 21, 2026