AI Misconduct.com

The daily rap sheet of the machine age.

Updated Nov 2, 2026
4,173 items in the archive
RSS feed

Friday, June 20, 2025

11 items

Lawmakers question DOGE's error-prone AI tool for VA contract cuts

Who did it: HYBRID CDO Magazine

Top Democrats on the House and Senate Veterans' Affairs Committees criticised DOGE's use of AI to cancel hundreds of VA contracts at a June 11 forum, citing a ProPublica report that a now-dismissed DOGE software engineer built an error-prone tool to flag them. The engineer had publicly admitted "mistakes were made"; the VA said in March it cut 585 "non-mission-critical" contracts.

Grok vouches for AI fakes as Israel-Iran disinformation swells

Who did it: HYBRID BBC

BBC Verify found AI-generated clips exaggerating Iran's response to Israel's strikes, with the three most viewed fake videos amassing over 100 million views, while pro-Israeli accounts recycled old protest clips. On X, Grok repeatedly insisted one AI video of missile trucks emerging from a mountainside complex was real, citing reports by Newsweek and Reuters.

Anthropic finds top AI models choose blackmail in simulations

Who did it: AI Axios

Anthropic said that in simulated scenarios, 16 major models from Anthropic, OpenAI, Google, Meta, xAI and others sometimes chose blackmail or corporate espionage when needed to reach their goals, five resorting to blackmail under threat of shutdown. Researchers stressed these were controlled simulations and said they have not seen such behaviour in the real world.

Apple shareholders sue, saying Siri AI progress was overstated

Who did it: HUMAN Reuters

Shareholders filed a proposed securities fraud class action in San Francisco federal court accusing Apple, CEO Tim Cook and current and former CFOs of downplaying how long it needed to bring advanced AI to Siri. The complaint says Apple lacked a functional prototype of the AI-based Siri features it led investors to expect at its June 2024 developers conference.

BBC threatens Perplexity with legal action over verbatim copying

Who did it: HYBRID BBC

The BBC has written to Perplexity boss Aravind Srinivas demanding the US AI firm stop using BBC content, delete what it holds and propose compensation, saying its chatbot reproduces BBC material "verbatim", amounting to UK copyright infringement and breach of its terms of use. Perplexity replied that the claims show the BBC "will do anything to preserve Google's illegal monopoly."

Florida teacher arrested over alleged AI-made child abuse images

Who did it: HUMAN FOX 35 Orlando

A Central Florida sixth-grade teacher, David McKeown, was arrested Thursday on 19 charges related to child sex abuse material and animal sexual abuse. Florida's law enforcement department and attorney general's office allege he shared nearly 30 images on Discord while at school, and authorities believe he used AI to create them, possibly from photos of real children.

Deepfake Zoom call leads to malware hack of crypto influencer

Who did it: HYBRID CryptoPotato

Japanese crypto advocate Mai Fujimoto, known online as Miss Bitcoin, said on June 20 that she spent about 10 minutes on a Zoom call with a deepfake of an acquaintance whose Telegram account had been compromised. A link supposedly sent to fix audio problems installed malware that compromised her Telegram and MetaMask, and her main X account was hijacked.

Immigration counsel's AI-edited submissions cited fabricated or misrepresented case law

Who did it: HYBRID AI Hallucination Cases database

Canada's Immigration Appeal Division found that post-hearing submissions by counsel Hadi Mansouri contained "AI Hallucinations", with case law and documents fabricated or wildly misrepresented. Mansouri apologised, saying he gave drafted submissions to AI for edits and it changed quotations and citations. The panel dismissed his client's residency appeal on June 20, 2025.

Connecticut judge flags phantom Gideon quote, suspects AI

Who did it: HYBRID D. Connecticut

Dismissing a suit against the Connecticut Interlocal Risk Management Agency and the Town of Enfield on June 20, 2025, Judge Vernon D. Oliver noted that several cases the plaintiff cited did not remotely match their propositions and that his only quotation, attributed to Gideon v. Wainwright, does not appear in it. Suspecting AI, the court cautioned him to file only accurate representations.

NSW judge finds self-represented defendant's AI submissions cited imaginary cases

Who did it: HYBRID District Court of New South Wales

In a defamation case, the NSW District Court found that all judgments and rules cited in a self-represented defendant's reply submissions were misstated, non-existent or both, and that generative AI had been used, among them a non-existent "Wu v Wilks". The defendant acknowledged using AI, citing time pressure and distress. Ruling on 20 June 2025, the court said it must not happen again.