Baidu apologises in China's first AI defamation case — two years on
A lawyer who searched his own name and found a criminal record that never existed finally got a written apology today. The reasoning behind the order is the part the industry should read.
Baidu formally apologised in writing to Nanjing lawyer Li Xiaoliang on September 22, at the Nanjing Jiangbei New Area People's Court — the first time a Chinese court has forced a model operator to apologise for what its model said about a named person. The case began in September 2024, when Li searched his own name on Baidu's app and website and found the "AI Smart Answer" panel telling readers that "lawyer Li Xiaoliang was sentenced to three years' imprisonment," for the crime of explosion, illustrated with a photograph of him in his barrister's robes. He complained for months — letters from counsel, Baidu's advertising hotline, a municipal government line — and each time Baidu replied that the matter had been passed to staff. The core claim stayed up.
Baidu's defence was the standard one, and the judge dismantled it with a comparison rather than a doctrine. The company argued that hallucination is unforeseeable, that every large model produces it, and that strict liability would deal a heavy blow to the industry; it invoked the service-provider protection of technical neutrality. According to the account published by the outlet 星河商业观察, the court asked Baidu why the same query put to Doubao and DeepSeek returned none of this content. That question is the whole ruling: if comparable products did not produce the defamation, the error was not the inevitable cost of the technology. Baidu had the ability to prevent it and did not, which is fault — and fault is what defamation law requires.
The procedural history is what gives the decision weight. The Jiangbei New Area court found for Li in September 2025 and ordered a written apology. Baidu appealed, and in March 2026 the Nanjing Intermediate People's Court rejected the appeal and affirmed. Li had also sought lost income and emotional damages; the courts declined, citing the technology's early stage, Baidu's partial remediation and insufficient evidence of loss — an apology without a payout. And the apology itself did not arrive on time. Li filed for compulsory enforcement after the deadline passed, which is how a one-sentence apology ended up requiring a bailiff.
Two more suits are already queued behind it: Beijing lawyer Huang Guigeng, whose name Baidu's AI attached to a list of crimes including threatening judges, brokering bribes and forging seals by substituting his name into unrelated news reports, filed in Haidian in November 2025 and is seeking 1 million yuan; writer Kou Yanding has filed a similar claim. Both are still being heard, and Baidu is running the same defence. China's courts have spent this year drawing exactly this line elsewhere — we covered the top court's rules for AI deepfake lawsuits and a ruling that AI customer service cannot promise refunds it can't deliver. The pattern across all three is the same, and it is narrower than "AI is liable": a platform escapes liability for a general risk of error it cannot eliminate, and does not escape it for a defamatory claim about a specific, identifiable person that a competitor's product did not make.
Ligent Technologies jumped as much as 19.2% in its Hong Kong debut on Tuesday, after the Qingdao optical-transceiver maker raised HK$5.67 billion, or about $723 million. The stock touched HK$39.30 against an offer price of HK$32.96 before settling at HK$35.90, up 8.9%, giving it a market value of roughly HK$32.4 billion at the offer price. The retail tranche was 35 times subscribed and the international book 4.67 times, with Primavera, GigaDevice, Amlogic's Hong Kong unit, Mirae Asset, PAG, Barings, GF Fund and E Fund among the cornerstone investors; parent Hisense Group retains about 40% control. The company said it will spend 52.9% of net proceeds on faster optical products and chips, and 25.1% on capacity. Its first-half net profit rose 29.7% to 661 million yuan on revenue of 5.39 billion yuan.
The reason a transceiver maker can raise that sum is that the bottleneck in AI data centres has moved off the accelerator. Once thousands of GPUs sit in one cluster, the chips are only as useful as the links between them, and those links are optics — a bill of materials tailwind we've seen in Lumentum's revenue doubling as AI data centres go optical. Shenton Research's Ke Yan makes the honest version of the bull case and the honest version of the risk in the same breath: vertical integration across optical module manufacturing is a genuine selling point, but Ligent's current shipments still skew toward the cheaper end of the product spectrum, and suppliers in this chain are priced for compute demand that keeps compounding. Any deceleration is a de-rating, not a pause.
What to watch: whether Baidu changes what its AI answer panel is allowed to generate about individuals before the Haidian ruling rather than after it, and whether Ligent's R&D spend moves it up the module stack before the pricing cycle turns.
If a platform has to be dragged to court and then to enforcement to correct a defamatory sentence about one named person, what does that leave for everyone who isn't a lawyer? Tell us in the comments.
