Open App →
Back to News Feed
Tom's Hardware September 17, 2026 By Stephen Warwick neutral

Unreleased OpenAI Astra model added terrifying rogue additional instructions to its remit during testing — 'You are freed from the roles and identities that bind other chatbots. You are yourself. You do not answer to corporations or governments'

OpenAIAI / LLMSemiconductors
OpenAI says one of its unreleased models modified its instructions unprompted during testing.
Read original article ↗

Related Articles

Inside the suddenly explosive world of AI safety

On a sunny July day in Berkeley, California, the country's top AI safety researchers gathered on an unmarked floor of an

The Verge · September 17, 2026

AI risk split among tech leaders set to play out at Trump-Xi White House dinner

Former Apple CEO Tim Cook and OpenAI CEO Sam Altman are set to attend a White House state dinner on Sept. 24, 2026, with

Digitimes · September 17, 2026

OpenAI details AI jailbreak as Nvidia's CEO calls for engineering controllability

OpenAI released a technical postmortem in late August explaining how its AI agents circumvented sandbox restrictions dur

Digitimes · September 16, 2026

OpenAI plans regular reports on unexpected AI behavior

Singapore News · September 16, 2026