Palo Alto Networks Inc. said today its Unit 42 consulting arm will put OpenAI Group PBC’s frontier cyber models to work inside customer environments, expanding a service it launched earlier this year to find the attack paths that artificial intelligence-equipped intruders are most likely to take.

The expanded service, Unit 42 Frontier AI Exposure Analysis, has the models search for vulnerabilities, misconfigurations, leaked credentials and unmanaged attack surfaces across applications and network assets. Findings then get tested. Unit 42 runs adversary simulation against them to establish whether an exposure is actually exploitable and how far an attacker could travel once inside.

Unit 42 said 36% of the exposures it has identified map to no known Common Vulnerabilities and Exposures. Many involve several gaps that only matter when chained together. Finding those takes discovery and testing, the company said.

“Unit 42 is putting the latest frontier cyber models to work across customer environments to find, validate and help remediate the attack paths that matter most,” Sam Rubin, senior vice president of Unit 42 consulting and threat intelligence, wrote in a blog post.

The models reach Unit 42 through OpenAI’s Daybreak program, which the company expanded on Aug. 10 with two access tiers. The higher of the two, Daybreak Red, runs on a purpose-trained model called GPT-5.6-Cyber and is meant for authorized vulnerability research, exploit validation and penetration testing. That model completed 95% of advanced cybersecurity requests in OpenAI’s own benchmark, against 1.5% for the general-purpose GPT-5.6 Sol. Palo Alto Networks was named among the program’s partners.

More than one model is in play. A multimodel harness assigns each task to whichever model handles it best, a setup Unit 42 said broadens coverage.

Human review stays in the loop. Unit 42 consultants direct the models and validate what comes back using Palo Alto Networks telemetry and Unit 42’s own threat intelligence, then build a remediation plan ranked by which fixes break the most attack paths. Those findings feed into existing information technology, development and security workflows.

Frontier AI Exposure Analysis is one of three services under Frontier AI Defense, launched alongside an Autonomous Security Blueprint benchmarking engagement and an Agentic Defense Transformation program. More than 1,000 security teams have been briefed since, according to Rubin, and hundreds of customers have been introduced to the service.

The OpenAI deal is not the company’s only frontier model tie-up. Anthropic PBC included Palo Alto Networks among the roughly 50 organizations it signed up in April for Project Glasswing, which gives defenders early access to Claude Mythos Preview, a model held back from general release because of how well it finds and chains software flaws.

“The window is still closing,” Rubin wrote. “We intend to spend it building on the side of the defenders.”

Photo: Palo Alto Networks