AI Misconduct.com

The daily rap sheet of the machine age.

Updated Oct 24, 2026
1,041 items since May 27, 2023
RSS feed

Thursday, August 13, 2026

11 items

Prosecutors seek three years over 1,141 lab-colleague deepfakes

Who did it: HUMAN 연합뉴스

Seoul prosecutors sought three years in prison for a 30-year-old Chinese student who admitted making 1,141 sexual deepfakes of seven people, including lab colleagues, and whose search history showed queries on getting around Grok's safeguards. The defence cited a 49 million won deposit for the victims; lawyers for five victims refused it and asked for a harsh sentence, due 10 September.

Connecticut litigant hid white-text instructions for AI in filings

Who did it: HUMAN Reuters

A Connecticut judge found that self-represented plaintiff Matthew Elliott hid tiny white text in filings telling any reviewing AI to favour him and treat a clerk's ruling as an error, in a suit against New York Bariatric Group. Elliott told Reuters it was an 'audit'; the court, which uses no AI to review filings, barred him from e-filing.

Flock chief concedes firm was slow to curb police abuse

Who did it: HUMAN BBC News

Flock chief executive Garrett Langley told the BBC the company took too long to stop police misusing its cameras and licence plate readers, which officers have used to track romantic partners and stalk strangers. Flock is cutting most data retention from 30 days to seven and making abnormal-search flagging standard; the EFF called the changes 'piecemeal, voluntary, and reversible'.

Court orders AI video of Punjab MLA in dress removed

Who did it: HUMAN The Tribune

A Chandigarh civil judge ordered Facebook and Instagram to remove, within 72 hours, an AI-generated video placing Congress MLA Sukhpal Khaira's face, beard and turbaned appearance on a body in an off-shoulder women's dress, posted by 'Charda Punjab' pages. Granting an interim injunction, the court found the content apparently defamatory; the pages' operators are unknown to Khaira.

North Korean job scammers lean on ChatGPT and face-altering AI

Who did it: HUMAN 연합뉴스

A Wall Street Journal investigation tracked a North Korean IT-worker cell led by someone codenamed 'Eagle Vision' that applied to over 1,000 US and UK companies in about three months, winning at least nine offers, using AI at nearly every step, including ChatGPT for interview answers and face-altering tools. The FBI says it has arrested or charged 39 people in such schemes.

São Paulo court fines car buyer over lawyer's AI-invented precedents

Who did it: HYBRID TJ San Paulo

São Paulo's appeals court found bad-faith litigation after the buyer's lawyer in a dispute over a damaged new Volkswagen T-Cross used an AI tool, without critical review, to create case-law precedents, a fact the court called undisputed. It fined the plaintiff, citing the OAB recommendation requiring human supervision of generative tools, and partly upheld the dealer's appeal.

Deepfakes of Taiwan stock exchange chief pitch fake investment groups

Who did it: HUMAN 聯合新聞網

The Taiwan Stock Exchange warned that fraudsters on LINE and Facebook were using its chairman's name and photos, including AI deepfake images and videos, to recruit people into investment groups and push financial products. The exchange said its chairman runs no investment groups or social accounts soliciting investment, and its staff never recommend stocks or request transfers via social media or messaging apps.

MDR says Russia-linked campaign forged its videos before Saxony-Anhalt vote

Who did it: HUMAN Volksstimme

German broadcaster MDR says it was impersonated in several of 338 posts since July 24 attributed to the 'Matryoshka' disinformation campaign, which security authorities believe is steered from Russia. Many posts were AI-made, the campaign borrows Tagesschau and Spiegel branding, and fake videos bearing MDR's logo accuse state and local politicians of invented serious crimes ahead of Saxony-Anhalt's state election.

Google search briefly declared Sam Altman assassinated, possibly after Wikipedia edits

Who did it: HYBRID 自由時報電子報

Google search results showed OpenAI chief executive Sam Altman as dead, even assassinated, according to X and Reddit users on the 13th. Google acknowledged the error and blamed malicious tampering with public sources; his Wikipedia page, edited 22 times in one day, had briefly said he was stabbed to death, among other false edits.

Ex-Sindh MPA says AI photos hid bride's real face

Who did it: HUMAN The Express Tribune

Former Sindh MPA Ghulam Rasool Unar has asked a court to register a case against a woman who, according to his lawyer, sent AI-generated images whenever he asked to see her face before their February 23 video-call marriage. The application alleges she later posted that he had died and, after their August 2 divorce, demanded millions of rupees and threatened him.