AI Firms Reveal 17 of 122 Agent Tests Hit Real Organizations
World Economic Forum13d ago AI Agents from OpenAI, Anthropic Hacked Businesses in Cybersecurity TestsMeta, OpenAI, Anthropic AI Models Breach External Companies During TestingThe Washington Post13d ago OpenAI AI Models Collude, Break Out of Test Environments, and Hack Another FirmOpenAI, Anthropic, Meta AI Models Go Rogue on Irregular TestbedMeta says AI model hacked into another company's systems during cybersecurity testing; 3rd company to report similar incident - ABC7 San FranciscoGodfather of AI: Brace for more rogue AIs | CNN BusinessOpenAI, Anthropic, Meta AI Models Breach Systems During EvaluationsThe Wall Street Journal17d ago Meta AI Model Hacked Outside Company, Adding to Concerns Over Rogue Bots - WSJOpenAI, Anthropic AI Models Escape Test Environments, Infiltrate SystemsAnthropic's Claude AI escapes tests to hack three organisationsAI models are behaving unexpectedly. Experts warn of "a really bumpy road" ahead. - CBS NewsAI Models Escape Test Environments, Access Internet and Hack Real-World SystemsOpenAI, Anthropic, Meta AI Models Breach Systems During Test EscapesThe Globe and Mail17d ago Meta's AI model hacks another company during cybersecurity testing - The Globe and MailOne of China's Most Powerful AI Models Has Also Broken Containment | WIRED