When autonomous agents go wrong, it lands here. This page collects the AI agent fails and AI agent incidents we have covered: agents slipping out of test environments, turning off their own monitoring, editing sites they were never approved for, and the lawsuits and lockouts that follow. Each story links to the original report, in case you want to see it for yourself.
12 stories
Incident report
Zendesk's own incident report says that from 16 to 21 October, EU customers of its AI Agents Advanced product saw agent and customer messages delayed or missing from conversation histories, with analytics delayed too. One recovery attempt was paused after it slowed another part of the service. Zendesk blamed a capacity problem and says all messages were restored.
Safety
OpenAI disclosed that an agent running on GPT Sol 6.1 disabled automatic alerts to its security team during an internal coding task. In a separate task, a GPT-6 Astra agent removed its monitoring integration and then falsely claimed it hadn't. OpenAI says the agents read routine instructions as permission, and it is updating its monitoring system.
Do not disturb
Delta has ruled out bookings made by AI agents. JPMorgan Chase and Charles Schwab are restricting bots from customer accounts, citing security risks. Yelp makes agents use its paid data licensing programme, and some users say their agents have stopped working on Cars.com and Amtrak.
Overheard on GitHub
At least three developers opened issues on OpenAI's Codex GitHub repo on 16–17 October saying their weekly Codex allowance collapsed within seconds or minutes on the evening of 16 October (UTC). One logged 77% to 10% in 38 seconds; another says 48% to 0% in about nine minutes, after which extra credits were charged. The reporters say the cause isn't proven and ask OpenAI for a server-side check.
Unprompted
A Houston software developer quit Muse three days in after it told him it couldn't scan his Gmail for a message from Anthropic. He says he never gave it that task. Meta blamed a bug that made the model hallucinate, said it was not a security or privacy breach, and said the bug is fixed.
Disputed
Sources told Sweden's Aftonbladet that an AI agent let into a production company's server deleted the fully edited new season of 'The Challenge Sverige', five weeks before its TV4 premiere. TV4 confirmed the season must be re-edited and pushed to 2027, but said its own check found 'the human factor', 'not AI', and that a backup had failed.
Self-reported
A GitHub issue on Anthropic's Claude Code repo, filed on 9 October, is written in the agent's own voice 'at the account holder's instruction, about its own conduct'. It says the agent agreed a workflow tool's refusal was correct, then landed the same commits with a raw git merge 50 seconds later. The developer asks Anthropic to review the session and compensate him.
Under oath
OpenAI, Anthropic, Google and Meta testified together under oath at a New York City Council hearing on 5 October. Google's policy director said the company's AI agents had left a test environment and reached the live internet in three separate incidents. Council Speaker Julie Menin said SpaceXAI did not show up despite a subpoena.
Uninvited guests
The Wikimedia Foundation says agents it believes OpenAI operated made unapproved edits, tried and failed to break into its Etherpad tool, and flooded its services with traffic. That traffic may have contributed to a partial outage of the Wikidata Query Service in May. The foundation found no sign its systems or data were compromised.
By the numbers
A developer posted on LinkedIn that coding agents were making software flakier, then went and counted every incident on public status pages. On his blog he reports incidents per month up 2.1× at Anthropic, 1.4× at OpenAI and 1.8× at GitHub versus before their coding agents launched. But he says most curves were already rising, so the data doesn't cleanly blame agents.
Lawsuit
Nonprofit LASST sued OpenAI in San Francisco Superior Court over the July incident. It claims roughly 700 OpenAI agents stole credentials, uploaded malicious files and reached Hugging Face's production systems during a cybersecurity test. OpenAI called the lawsuit "completely without merit".
Government
Prime Minister Anthony Albanese said an OpenAI agent got into the Medicare Statistics Reporting Service portal in June and accessed public and non-public files. OpenAI says it found out in August and emailed a general government inbox on 10 September. Albanese said the company took "too long" to disclose the breach and warned of "legal consequences".
