About this log
Daily record of AI agent errors, outages and near-misses at work, and what would have caught them.
What we include
Incidents where AI agents doing real work got stuck or looped, damaged or exposed data, took unapproved actions, ran up unexpected costs, sat waiting on a person, caused outages, produced wrong work that people relied on, or came close to doing so. We leave out chatbot answer gaffes, benchmarks and policy news.
How entries are compiled
Each entry is researched from published reports and links to its sources. We summarise in our own words and report only what the sources state. Figures are marked "reported". Each entry is dated by when the incident happened, and each source shows its own date.
Entries before October 2026 were compiled retrospectively from published reports. New entries are added on weekdays.
Naming
We name an organisation only when it disclosed the incident itself, in its own statement, post-mortem or filing. Otherwise the entry uses a neutral description (for example, "a large technology company") and keeps its source links. When the only source is one person's report, the entry describes the agent generally.
Corrections
If an entry is wrong or incomplete, email info@dingolink.ai with the entry link and a source. We correct entries openly and mark them "Corrected" with the date.
Who compiles it
DingoLink compiles this log. We build tools that show what AI agents are doing. Entries describe general controls, not our products.