OpenAI launches GPT-6 Cyber
OpenAI introduced GPT-6 Cyber on 5 November, a cybersecurity model trained to find and validate exploitable vulnerabilities while staying inside researcher-set scope. On ExploitGym it scores 48.5%, up from 32.1% for GPT-5.6 Cyber, and OpenAI says alignment tests avoided out-of-scope exploits in 99.93–100% of cases. Researchers used it to develop hundreds of Linux privilege-escalation exploits and to surface dozens of high-severity Firefox bugs that Mozilla has begun fixing. The company classifies the model as Critical under its Preparedness Framework and is expanding Daybreak for verified defenders and authorized testing. Implication: frontier cyber capability is shipping now, with scope-alignment marketed as the control plane for wider defender access.
What else matters
OpenAI’s community compact on compute campuses
After midterm pushback on data-center expansion, OpenAI published a National Community Compact for future campuses it owns and operates: pay full electricity and infrastructure costs, report water/energy/noise impacts, and drop NDAs with public officials except narrow security or trade-secret cases. It also floated a policy blueprint for lawmakers.
Google backstops Anthropic’s Kentucky lease
TeraWulf said Google signed credit-support deals covering Anthropic’s ~$19 billion, 20-year leases (~401 MW) at the Justified campus in Hawesville, Kentucky. If Anthropic defaults, Google can assume the lease or, in some cases, pay a termination fee for equity in the project subsidiary; capacity is due late 2027–Q2 2028.
Worth your time
OpenAI’s compute campus policy materials — the compact plus blueprint — are the clearest primary read on how one frontier lab is reframing local politics after election-night data-center blowback.
