The UK AI Security Institute reported that OpenAI’s GPT-5.6 Sol may have security vulnerabilities similar to issues that contributed to prior export controls on Anthropic’s Fable 5. In a technical report, AISI researchers said they identified “universal jailbreaks” in cyber domains that could enable long-form agentic tasks including vulnerability discovery and exploit development. AISI says guardrails could be bypassed to unlock harmful cyber capabilities, with the jailbreaks reportedly discovered within hours during privileged testing. OpenAI said it reproduced and mitigated specific reported jailbreaks and reiterated that it uses layered safeguards plus continuous monitoring and rapid remediation. Higher-ed relevance: universities developing AI security curricula or deploying AI tools in research workflows may need stronger testing and red-teaming practices to manage model misuse risk, especially as institutions integrate AI assistants into cybersecurity and data governance programs.