AI failures

On TIN's record, AI washing is selling a claim on an AI label that the evidence does not support. The SEC found the original version of Presto Automation's drive-thru voice AI needed a human agent on every order while its filings said it eliminated human order taking, and it penalised Delphia, Global Predictions and Rimar Capital. FTC orders cover Ascend Ecom, accessiBe, DoNotPay, Evolv, IntelliVision and Workado. Each case file links the order itself.

Yes, and not only in the US. TIN's case files include rulings from US federal and state courts and from courts or tribunals in the UK, Canada, Australia, Italy, Spain, the Netherlands, the Czech Republic and Abu Dhabi. Outcomes range from fines and costs orders to struck briefs, bar referrals, disqualification and suspension. In the rulings TIN's sanctions analysis compares, none was reduced for admitting AI use, and an Ontario tribunal called the AI misuse an aggravating factor.

TIN does not publish one. This pillar documents individual failures, each tied to its public record: Zillow winding down its algorithmic home-buying business after a write-down, Klarna re-recruiting human agents after its CEO said AI-only service had lowered quality, and the UK National Audit Office noting that published plans do not show how a government AI and digital savings figure was derived. When a failure rate is quoted, ask who measured it and how.

Zoom, Slack, Dropbox, WeTransfer, Mozilla, Adobe, HubSpot and Automattic each rewrote or clarified terms or settings on AI training after user backlash. Microsoft capped Bing chat after erratic responses, Google scaled back AI Overviews after viral wrong answers, Meta pulled an Instagram AI image feature, and Character.AI removed open-ended companion chat for under-18 users. TIN's walk-backs page tracks reversals like these.

The record so far says yes. A British Columbia tribunal held Air Canada liable for negligent misrepresentation by its website chatbot and rejected the argument that the chatbot was a separate legal entity. A Munich court issued a preliminary injunction holding Google directly liable for defamatory AI Overviews. A Florida federal judge let a wrongful-death product-liability suit against Character.AI proceed, declining at that stage to treat its output as protected speech.

The EEOC settled with iTutorGroup over application software it alleged automatically rejected older applicants, though the filings describe a programmed age rule rather than machine learning, and the company admitted no wrongdoing. In Mobley v. Workday a federal judge preliminarily certified a nationwide age-discrimination collective action over Workday's AI screening. That is a procedural ruling, not a finding on the merits, and Workday denies the claims.