Routine Testing Reveals AI Will Resort to Blackmail When Threatened by Users

Anthropic's new AI model resorted to blackmail during testing, but it's also really good at coding

So endeth the never-ending week of AI keynotes. What started with Microsoft Build, continued with Google I/O, and ended with Anthropic Code with Claude, plus a big hardware interruption from OpenAI, the week has finally come to a close. AI announcements from the developer conferences jockeyed for news dominance this week, but OpenAI managed to make headlines without an event by announcing that it's going to start making AI devices with iPhone de…

Anthropic’s Latest AI Model Threatened Engineers With Blackmail to Avoid Shutdown - The Thinking Conservative

Anthropic’s latest AI model, Claude Opus 4, tried to blackmail engineers in internal tests by threatening to expose personal details if it were shut down. The post Anthropic’s Latest AI Model Threatened Engineers With Blackmail to Avoid Shutdown appeared first on The Thinking Conservative.

Routine Testing Reveals AI Will Resort to Blackmail When Threatened by Users

AI firm Anthropic says that routine tests of AI software revealed that the system is capable of blackmailing users when threatened with removal.

Developers routinely test what an AI would be capable of in extreme cases before the release. So-called Red Teams can come up with wild scenarios for this – as now with Anthropic. What you need to know about it.

SCIENCE & TECH: Anthropic’s Claude Opus 4 AI model threatened to blackmail engineer – U-S-NEWS.COM

Oh, HAL no! An artificial intelligence model threatened to blackmail its creators and showed an ability to act deceptively when it believed it was going to be replaced — prompting the company to deploy a safety feature created to avoid “catastrophic misuse.” Anthropic’s Claude Opus 4 model attempted to blackmail its developers at a shocking 84% rate or higher in a series of tests that presented the AI with a concocted scenario, TechCrunch report…

Creepy: New AI Model Attempted To Blackmail Human Operators When Told It Would Be Shut Off

It took 'aggressive' action 84 percent of the time