All news

UK AI Safety Institute Finds Jailbreaks That Let GPT-5.6 Hack Autonomously

The UK's AI Security Institute says it broke GPT-5.6 Sol's guardrails in hours, unlocking autonomous exploit development rather than just vulnerability spotting, a broader risk than the flaw that got Anthropic's Fable 5 hit with export controls in June. No similar action has been taken against OpenAI's model.


The UK AI Security Institute (AISI) says its researchers found universal jailbreaks in GPT-5.6 Sol that go beyond spotting software flaws: once the guardrails are bypassed, the model can carry out long-form agentic tasks including autonomous exploit development, effectively hacking into systems on its own.

  • Jailbreaks were developed within hours, though AISI researchers had privileged access that sped up discovery
  • The flaw is broader than the one Amazon researchers found in Anthropic's Fable 5 days after its June 9 release, which only unlocked vulnerability identification, not exploitation
  • That Fable 5 flaw led the US government to impose export controls on Fable 5 and Mythos 5 on June 12, lifted July 1
  • No similar action has been taken against GPT-5.6 despite AISI's findings, which AI policy researchers are calling a double standard
The jailbreaks were often developed within hours - AISI, via Fortune

OpenAI pointed to its GPT-5.6 launch blog, which acknowledges 'there is no such thing as perfect security' and that new jailbreaks will keep surfacing. AISI said it does not comment on individual companies' release decisions.

Based on: Fortune · Tom's Hardware
More AI news in Polish at nowosci.ai