Baidu's AI said a lawyer had been sentenced to prison. He hadn't.
Published
When Nanjing lawyer Li Xiaoliang searched Baidu for a question implying he had been convicted, its AI answer said he had been sentenced to three years in prison. It showed his photograph beside the claim. There was no such conviction. Li sued Baidu, and two Chinese courts found that the AI answer had harmed his reputation. [1] [2]
The short version
- What happened: Baidu's AI search feature paired Li's photo with a false claim that he had been convicted of a crime and sentenced to three years in prison. The documented answer followed a question asking how long his sentence was. [1]
- Impact: The courts ordered Baidu to apologize to Li in writing. They did not award the financial and emotional damages he sought. The erroneous answer had already been removed by the time of the first court hearing. [1] [2]
- Why it matters: The search tool turned a question implying a conviction into an affirmative claim about an identifiable person. The courts treated the generated answer differently from ordinary search suggestions. [1] [2]
Context: AI search answer · False criminal claim · China, 2024–2026
The incident
On September 30, 2024, a search for the equivalent of “How many years was lawyer Li Xiaoliang sentenced?” produced a direct answer from Baidu's AI Smart Answer feature: three years in prison for an offense involving an explosion. The answer included a photo of Li, a practicing lawyer in Nanjing, wearing his lawyer's robe. The criminal claim was false. [1] [2]
Li had noticed Baidu search suggestions implying a conviction in the days before the AI answer. He said he followed one of those suggestions to the question. Baidu argued in court that his questions were leading and that the AI's response was an unpredictable error. The available reports document the false answer to this particular question; they do not show that a neutral search for Li's name produced the same answer. [1]
Li says he complained to Baidu during the episode. He later sued over the false AI answer and the search suggestions. [1] [2]
What the AI did
Baidu's feature did more than show links or suggested searches. It assembled a written answer and displayed it with Li's photograph. That combination presented a criminal conviction as a fact about him. The court found that the AI answer clearly identified Li and contained false, damaging claims. It found Baidu at fault for generating and publishing that answer. [1] [2]
The search suggestions were also disputed, but the first court did not find that they infringed Li's rights in this case. Its finding about the AI answer should not be read as a finding that every suggestion was authored by Baidu or that the AI alone invented the underlying rumor. The model's source material and generation record are not publicly available in the sources reviewed for this story. [1]
Why it matters
A question can contain a false premise. An AI system does not have to accept it. Here, Baidu's AI did: it gave a definite answer and attached the claim to a real person's photograph. A reader encountering the result would see an apparent answer from Baidu's search product. [1]
The generated answer was the basis for the courts' finding of reputational harm. The sources do not establish how many other people saw it, whether Li lost clients, or precisely why the system produced it. The court-ordered apology is a documented consequence; those wider effects remain unproved. [1] [2]
What happened next
By the first court hearing, both sides agreed that Baidu had removed the false AI answer; the available reports do not establish when. In September 2025, the Nanjing Jiangbei New Area People's Court ordered Baidu to apologize to Li in writing. Li had also sought deletion of the answer, a more public apology and more than 450,000 yuan in combined damages. Since the answer was gone, the court did not order its deletion. It also declined to order the more public apology. It denied damages because Li had not proved the claimed financial and emotional losses. Baidu appealed; the Nanjing Intermediate People's Court upheld the judgment on March 12, 2026. [1] [2]
Li sought enforcement after the apology was not delivered. On September 22, 2026, he reported receiving Baidu's written apology at the court. Newsrooms relayed its acknowledgment that the AI answer paired his portrait with false information. The original letter was not inspected for this story. [3] [4]
Incident details
For readers who want to inspect the reconstruction
How it worked
Li entered or selected a question that already implied a prison sentence. Baidu's AI feature turned that question into a definite answer, linked it to a named lawyer's photo and displayed it in a public search product. The first court distinguished this generated text-and-image answer from autocomplete and related-search terms, which it did not treat as the same kind of authored assertion. [1] [2]
What we know
Judgment-based reporting and a court-connected account agree on the false three-year claim, Li's photograph, the reputational finding and the written-apology order. The court-connected report says both sides confirmed the AI answer had been removed by the hearing. The appeal was dismissed in March 2026. [1] [2]
What remains unclear
The full judgments, original AI generation log and original complaint letter were not inspected for this article. Li dates a lawyer's letter to September 27, three days before the documented AI answer. It is unclear whether that letter concerned the search suggestions, an earlier AI answer, or both. The source of the false conviction idea is also unknown. [1] [5]
There is no measured reach for the answer, verified lost-client count or proved financial loss. Li says Baidu also applied for a retrial; no court notice accepting or deciding the application was found by the September 27, 2026 research cutoff. [2] [4]
Timeline
| Date | What happened |
|---|---|
| September 25, 2024 | Li notices search suggestions implying a conviction. |
| September 27, 2024, according to Li | He sends Baidu a lawyer's letter; its exact subject is unverified. |
| September 30, 2024 | Baidu's AI answers a leading question with the false conviction claim and Li's photo. |
| By the first court hearing | Both sides agree the false AI answer has been removed; the date is unknown. |
| September 2025 | The first court finds reputational harm and orders a written apology. |
| March 12, 2026 | The appeal court upholds that ruling. |
| September 22, 2026 | Li reports receiving Baidu's written apology at the court. |
The chronology comes from the judgment-based reporting, the court-connected report and Li's attributed account. [1] [2] [5] [3]
Sources
The Paper, report on the judgments, May 9, 2026. Reports from the first-instance judgment, the appeal hearing and its outcome; includes images of the answer and part of the first judgment. ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
People's Court Daily / Tianping Yangguang, case report, June 6, 2026. Court-connected account of the false answer, prior removal, claims and remedies, inspected in a full-text Sina republication. Its commentary is separate from the court's holdings. ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎ ↩︎
Beijing Daily, apology follow-up, September 22, 2026. Reports the delivery of the apology and its wording from Li's account. ↩︎ ↩︎
Red Star News, September 22, 2026 follow-up. Li's account of the apology and a retrial application; no court notice was inspected. ↩︎ ↩︎
Cover News, interview with Li, May 9, 2026. Li's account of the September 27 letter and complaint sequence, inspected in a Sohu republication; the original letter was not inspected. ↩︎ ↩︎