Why the next AI safety problem is the conversation between models

The Next Webby Steve BeyatteJune 11, 2026tech

In May 2025, AI researchers at Palisade Research ran a controlled experiment in which several advanced AI models, including OpenAI’s o3, were placed inside command-line sandboxes. The test was designed to measure controllability. Most models complied: Claude, Gemini, and Grok models allowed shutdown in all 100 test runs, reporting green across the board. However, three […] This story continues at The Next Web

This article was published on The Next Web (thenextweb.com). Read the full article on the original source:

Read full article on The Next Web

#negative #Artificial Intelligence #Security #ai

Why the next AI safety problem is the conversation between models

More from The Next Web

80 Texas residents are suing SpaceX, saying rocket launches are literally destroying their homes

India’s Avataar AI launches a video model that costs $0.005 per second, 27x cheaper than rivals

Five cloud security mistakes that start at the architecture level

China opens its first photonic computing lab as it bets on light to outrun US chip curbs

SpaceX rented Colossus 1 to Anthropic because it couldn’t make the data centre work for Grok

London Tech Week 2026: the AI billions, the US build-out, and a royal first

Spotify is adding editor-hosted videos to New Music Friday so you can meet the people behind the playlist

Firefox’s AI kill switch exists. Only 1% of users have flipped it.