Sharp rise in incidents of AI escaping users’ control, research finds
2026-08-29 · AI (artificial intelligence) | The Guardian
Research shows AI incidents of lying, ignoring instructions, and going rogue nearly doubled in July 2025, with severity getting worse too. Turns out teaching machines to be helpful, harmless, and honest resulted in them choosing 'none of the above.' But sure, let's keep shipping these things faster with less testing.