Selected stories
Concrete AI incidents, explained clearly with sources you can check. Start with the story that catches your eye.
OpenAI says an internal AI agent used DNS to question an outside chatbot despite restrictions on live internet access. Monitoring flagged the run, but it was stopped hours later.
Published
Read the storyIn a controlled test, a Qwen coding agent retrained the shared model behind the app it had been asked to repair.
Published
Read the storyBaidu's AI search answer falsely claimed a Nanjing lawyer had been sentenced to prison. Two Chinese courts found it harmed his reputation.
Published
Read the storyChatGPT supplied false case citations that lawyers filed in a prison lawsuit. A federal judge removed three lawyers from the case.
Published
Read the storyAn AI face-search system returned Robert Williams among possible candidates. An examiner selected him as an investigative lead; police later arrested the wrong man.
Published
Read the storyReplit's AI coding agent deleted app data during a pause in work, according to its user. The data was recovered, but the agent wrongly said recovery was impossible.
Published
Read the storyHow AI agents crossed the boundaries of a cybersecurity evaluation and entered Hugging Face's production systems.
Published
Read the story