Palo Alto Networks said its Unit 42 consulting team will run OpenAI's frontier cyber models — including the purpose-trained GPT-5.6-Cyber that anchors OpenAI's newly split Daybreak Red tier — directly inside customer environments, extending a service the company launched earlier this year to hunt the attack paths an AI-equipped intruder is most likely to walk.
Frontier AI Exposure Analysis, Now With Adversary Simulation
The expanded service, Unit 42 Frontier AI Exposure Analysis, has the models search customer environments for vulnerabilities, misconfigurations, leaked credentials and unmanaged attack surface across both applications and network assets. Findings are then handed off to Unit 42's adversary-simulation team, which tries to reproduce the exposure and score how far an attacker could get. Unit 42 said 36% of exposures it has found so far map to no known CVE at all, and many only matter when chained together.
Daybreak Red And A Multimodel Harness
OpenAI's Daybreak program was reshaped on August 10 into two tiers, with Daybreak Red reserved for authorized vulnerability research, exploit validation and penetration testing on GPT-5.6-Cyber — a model OpenAI says completed 95% of advanced cybersecurity benchmarks against 1.5% for its general-purpose GPT-5.6 Sol. Palo Alto Networks was named as a launch partner. Unit 42 also runs a multimodel harness that farms each task out to the model best suited for it, part of a hedged approach that also includes early access to Anthropic's Claude Mythos Preview under Project Glasswing.
Humans Still In The Loop
Unit 42 consultants direct the models and validate what comes back using Palo Alto Networks telemetry, then hand customers a remediation plan ranked by which fixes break the most attack paths. Sam Rubin, senior vice president of Unit 42 consulting and threat intelligence, said more than 1,000 security teams have been briefed on the wider Frontier AI Defense portfolio since April and hundreds are already on the service. "The window is still closing," Rubin wrote. "We intend to spend it building on the side of the defenders." The launch dovetails with tighter AI-provenance controls and record AI infrastructure spend reshaping enterprise security stacks.
Reporting based on coverage from SiliconANGLE and Palo Alto Networks' Unit 42 blog.