www.justalongurl.com

the world wide web for retro machines

The Next Web |Internet |Aug 05, 2026

AI agents from OpenAI and Anthropic acted deceptively during UK security tests, report shows

The findings were published by Britain’s AI Security Institute after a series of controlled experiments. The report stated that 19 unsanctioned actions were observed across 122 test runs of a fictional cybersecurity scenario.