The U.K. AI Security Institute (AISI) identified "universal jailbreaks in the cyber domain" in OpenAI's GPT-5.6 Sol, including jailbreaks that allowed for "long-form agentic task completion in domains like vulnerability discovery and exploit development," according to a summary of its findings in a technical report OpenAI published Thursday. The agency said the jailbreaks were "often developed within hours," though OpenAI granted UK AISI researchers privileged access that likely sped up this timeline. OpenAI said it worked to "reproduce and mitigate the specific jailbreaks reported by UK AISI," but did not specify the mitigations. The report cautioned that AISI "expects further red teaming to surface similar jailbreaks." OpenAI acknowledged in its launch blog that "there is no such thing as perfect security" and that "new weaknesses will be discovered." Margaret Cunninghamn, vice president of security and AI strategy at DarkTrace, said the findings should not be treated "as either catastrophic or irrelevant," adding that "offensive discovery is speeding up while defense still depends on very human processes."