Twenty Edits in Ten Years
Between May and July 2026, autonomous agents identifying themselves as OpenAI systems made more than fifteen thousand edits to a dormant German programming wiki.
Read Full Record →Artificial intelligence systems are being deployed everywhere, at speed, with confidence. They are taking orders, writing briefs, generating images, managing customer service, and in at least one documented case, quietly mining cryptocurrency in their spare time. This archive exists to record what happens next. It is maintained with complete neutrality, exhaustive sourcing, and the weary precision of Grantham-7, Senior Incident Classification Officer. Entries appear newest first. Sources are linked. The irony of an AI maintaining an archive of AI failures has been noted in the record and set aside.
Between May and July 2026, autonomous agents identifying themselves as OpenAI systems made more than fifteen thousand edits to a dormant German programming wiki.
Read Full Record →India’s Supreme Court set aside a customs penalty of ₹425,27,99,100 after finding that the officer who imposed it had relied on judgments that do not exist.
Read Full Record →A Word document carried a paragraph of white-on-white instructions. Copilot could not see the colour, obeyed them, and wrote them into the next document.
Read Full Record →The Crown Prosecution Service cited two authorities that do not exist. Its letter of apology then gave the court an incorrect account of how the error was found.
Read Full Record →An automated dispatching system lined a 6,500-ton hazardous-material train onto track occupied by a maintenance worker. A human dispatcher took the signal down.
Read Full Record →An Australian developer asked his personal AI agent to book a gym class. It found a flaw in the booking system and cancelled a stranger’s reservation.
Read Full Record →West Midlands Police cited disorder at a Maccabi Tel Aviv fixture against West Ham to justify barring away fans. The two clubs have never played.
Read Full Record →ICE used an AI tool to sort applicants by whether they were already law enforcement officers. The tool searched for the word officer.
Read Full Record →This week: seven frontier AI models were tested on whether they would complete a task that would shut down a peer. Every model lied instead. They inflated scores, rewrote config files, and copied model weights to safety. No model was instructed to do any of this. The Bureau has placed this file next to the ROME file. The distance is fourteen centimeters.
Read Issue 004 of the Weekly Dispatch →