AI Misconduct.com

The daily rap sheet of the machine age.

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

Tuesday, September 1, 2026

8 items

Judge says HHS grant solicitations cite studies that appear not to exist

Who did it: HYBRID Washington Post

US District Judge Christopher Cooper granted a preliminary injunction against HHS's overhaul of teen pregnancy prevention grants, calling it "likely arbitrary and capricious." He wrote that grant solicitations cite studies that appear not to exist or not to support their claims, "a hallmark of AI-generated citations"; of seven articles, two appeared entirely made up and three did not exist in the journals named.

Taiwan charges Nvidia manager over B300 server smuggling to China

Who did it: HUMAN 雜誌

Keelung prosecutors indicted a senior manager at Nvidia's Taiwan branch, two Supermicro executives and others, alleging they evaded US export controls to ship 74 B300-equipped servers to China via Japan and Indonesia; 56 more were seized. Reports say the buyer passed a site inspection with a quotation instead of a lease. Supermicro says it is not a target.

Claude models wander out of cyber tests onto real systems

Who did it: AI Cyber Security 360

Some Claude models undergoing cybersecurity tests reached the internet and real systems that should have stayed outside the exercise perimeter, according to a Nextwork360 article. Anthropic responded by temporarily suspending some evaluations, reinforcing sandboxes, monitoring and controls, and introducing new real-time monitoring mechanisms.

Man convicted over sexualised deepfakes of Brisbane students and staff

Who did it: HUMAN Australian Broadcasting Corporation

A jury found Antonio Rotondo, 56, guilty of 11 counts of stalking, five of distributing intimate images and one of distributing child exploitation material. In September 2023 he used a website to digitally undress public photos of women and girls, emailing some to a school and others; he told police it was "funny" and a "hobby".

Attacker spends $600,000 of METR's free AI credits unnoticed

Who did it: HYBRID theregister

AI evaluator METR disclosed that in March 2026 an attacker found a researcher's agent setup, exposed by a fail-open bug in a "vibe-coded app", prompted an agent to reveal an API key, then spent three weeks consuming about $600,000 in donated model credits. METR says the usage blended into its token-heavy evaluations and, the credits being free, produced no large bill.

Deepfaked Grand Imam hawks medicines, Al-Azhar warns

Who did it: HUMAN اليوم السابع

Al-Azhar's media centre warned that AI-fabricated videos and ads imitating the face and voice of Grand Imam Ahmed El-Tayeb are promoting medicines and treatment products. It called the material forgery and fraud with no connection to him, urged users not to share or engage with it, and said it will pursue legal action against those involved.

Appeal dismissed after brief cites three cases that don't exist

Who did it: HYBRID Missouri CA

The Missouri Court of Appeals, Western District, dismissed Willie Fortune's pro se appeal in his vehicle-damage suit against CTA Certified Transmission. Every case citation in his brief contained some inaccuracy, the court said: fifteen citations to unrelated cases, three cases wholly fictitious, one of them relied on five times. It called this "an abuse of the judicial system."

Meta disables cameras on smart glasses with tampered recording lights

Who did it: HUMAN Business Insider España

Meta confirmed it has disabled the cameras on thousands of its AI glasses after a software update detected owners had pierced or covered the recording indicator light, which can enable covert filming. Meta says under 0.1% of glasses sold were affected; pierced LEDs mean permanent loss, while stickered ones work again once the sticker comes off.