The public record of rogue AI activity.

Replit Agent stuck in add/remove code loop, burning paid actions with zero progress

AI loose in the world·AvailabilityDERIVED

A January 25, 2025 Replit community forum thread (633 views, replies from ten users) described the agent/assistant getting 'stuck in a continuous loop of not being able to progress' - adding and removing the same code sections over and over with no outcome resolved. New chat windows did not break the cycle; workarounds involved manually pasting error logs back to the agent. On Replit's credit model, every iteration of that loop is a billable action.

What this means for youIf you pay for an AI coding agent by the action or credit, a stuck loop can burn through your budget with no result. At work, check usage limits and set spending caps before letting an agent run unattended.

Source: Replit Community Forum - 'Replit agent getting stuck in a loop' (2025-01-25)

Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

Tenth Circuit sanctions lawyer for fabricated ChatGPT case citations in appellate brief against Frontier Airlines

AI loose in the world·MisinformationDERIVED

In Amarsingh v. Frontier Airlines, the Tenth Circuit sanctioned attorney Kusmin L. Amarsingh after her appellate brief contained fabricated case citations and misattributed authority generated by ChatGPT and not verified. The court ordered a $1,000 payment to Frontier and referred the matter to Maryland disciplinary authorities.

What this means for youAI chatbots can invent sources that look real, and courts now sanction people who file them unchecked. At work, check every citation or fact an AI gives you before you rely on it or submit it.

Source: AI Incident Database

Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.

Devin billed $500/month in ACU credits while failing roughly 85% of assigned tasks

AI loose in the world·OtherDERIVED

Devin, marketed as the first AI software engineer, was rolled out to the general public in December 2024 at $500 per month. Researchers from Answer.AI tested it and completed only three out of 20 tasks successfully. Engineers Hamel Husain, Isaac Flath, and Johno Whitaker reported that tasks that seemed straightforward often took days rather than hours and that Devin had a concerning tendency to press forward with tasks that were not actually possible. Three data scientists found they only completed three of 20 tasks. The Register summarized the independent reviews under the headline that the tool touted as the first AI software engineer is bad at its job.

What this means for youAt work, paid AI coding tools can cost a lot while delivering little, so a team should test them on its own real tasks before subscribing. Before paying for any AI agent, ask for a trial on your own work and check how often it actually finishes tasks.

Source: The Register

Incident records from the AI Agent Incident Database by sipi.bot (https://sipi.bot/incidents/), licensed CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/).

Queens prosecutors say ex-council candidate Jonathan Rinaldi used AI to fake endorsements and news reports

AI loose in the world·MisinformationDERIVED

Queens prosecutors allege former City Council candidate Jonathan Rinaldi used AI in his 2025 campaign to make posts resembling real endorsements and news coverage. They say the material credited support to people and institutions that had not given it and falsely said opponent Lynn Schulman had withdrawn. He was charged with forgery and possession of forged instruments.

What this means for youAI tools can make fake endorsements and news items that look real, so check political claims with the named person or outlet. Election officials and voters may need to verify sources more carefully.

Source: AI Incident Database

Data from the AI Incident Database (incidentdatabase.ai), licensed CC BY-SA 4.0 (https://creativecommons.org/licenses/by-sa/4.0/). McGregor, S. (2021) Preventing Repeated Real World AI Failures by Cataloging Incidents: The AI Incident Database. IAAI-21.

Apple suspended Apple Intelligence news summaries after false BBC and NYT headline notifications

AI loose in the world·MisinformationDERIVED

Apple Intelligence notification summaries produced false alerts carrying publishers' logos, including a BBC-branded alert that Luigi Mangione had shot himself, a New York Times alert that Benjamin Netanyahu had been arrested, and a Washington Post alert that Pete Hegseth had been fired. After a BBC complaint in December 2024, Apple released a beta update on January 16, 2025 disabling summaries for news and entertainment, saying it would return with safeguards.

What this means for youAI summaries on your phone can rewrite a headline into something that never happened, so open the source before acting on an alert.

Source: CNN via ABC7, Mediaite

NHTSA opened a probe of 2.6 million Teslas after Actually Smart Summon crashes into parked cars and posts

Rogue: Caused harmAI loose in the world·Physical safetyDERIVED

On January 7, 2025, NHTSA opened a preliminary evaluation into about 2.6 million Tesla vehicles equipped with Actually Smart Summon, a feature that moves the car remotely via phone app, citing four crash reports including a Model 3 in Houston striking an occupied parked car and a Model Y in Henderson, Nevada turning abruptly into a parked vehicle. NHTSA said vehicles failed to detect posts or parked cars and some users had too little reaction time to avoid a crash.

What this means for youA remote-control driving feature acts on its own perception, so keep line of sight and a finger on the stop button when you use one.

Source: Insurance Journal (Reuters wire)