Palo Alto Networks Inc. said its Unit 42 consulting arm will begin using frontier cyber models developed by OpenAI Group PBC directly in customer environments, expanding a service launched earlier this year to identify potential attack vectors for AI-enabled hackers.
„Unit 42 Frontier AI Exposure Analysis is an advanced analytics platform that scans for vulnerabilities, misconfigurations, leaked credentials, and unmanaged attack surfaces at both the application and network levels. The collected data is then validated by Unit 42 running adversary simulations to assess whether a detected vulnerability is actually being exploited and how far an attacker could go. This practice allows you to not only see the threat, but also assess its potential impact.
„How Frontier AI Exposure Analysis works
At the heart of the service is the OpenAI Daybreak program, which Palo Alto Networks expanded on August 10, 2026, with two access levels. The higher level, Daybreak Red, runs a specially trained GPT‑5.6‑Cyber model for authorized vulnerability scanning, operational verification, and penetration testing. According to OpenAI’s own tests, this model completed 95 % of complex cybersecurity queries, compared to 1.5 % of results for the general-purpose GPT‑5.6 Sol model.
The models operate in a multi-model framework – each task is automatically assigned to the model that best performs it. This allows for increased coverage and accuracy. Despite the automation, people remain important: Unit 42 consultants run the models, validate the results using Palo Alto Networks telemetry and their own threat intelligence, and then develop remediation plans that prioritize minimizing potential attack paths.
Analysis and practical application of results
Unit 42 says that 36 of the vulnerabilities discovered by % do not match any known Common Vulnerabilities and Exposures (CVE) entries. Most of these vulnerabilities involve a chain of multiple flaws that only become dangerous when they come together. Detecting such a chain requires not just a simple scan, but also a detailed testing phase, in which models and humans perform the same tasks.
The collected data is integrated into existing IT, development and security processes, allowing organizations to quickly implement recommended fixes. Remediation plans are sorted by which fixes break the most attack paths, giving the security team a clear set of action priorities.
Other frontier models and partnerships
The OpenAI deal isn’t Palo Alto Networks„ only frontier model partnership. The company is also participating in Anthropic PBC’s Glasswing project, where it will get early access to Claude Mythos Preview, a model that is exceptionally good at finding and patching software flaws. The partnership shows that Palo Alto Networks is looking to leverage several advanced AI tools to enhance its defense capabilities.
Service acceptance and future prospects
Since the service’s launch, more than 1,000 security teams have been informed and hundreds of customers have already become familiar with Frontier AI Exposure Analysis, demonstrating the strong interest and potential in a market where organizations are constantly looking for effective tools to predict and neutralize cyber threats.
„The window is still closing,“ wrote Sam Rubin, senior vice president at Unit 42, about the potential for these models. „We intend to spend it building on the side of the defenders.“ This quote emphasizes that Palo Alto Networks’ goal is to provide defense-side professionals with tools that will help them stay one step ahead of potential attackers.
Conclusion
Palo Alto Networks and OpenAI’s collaboration marks a new era in cybersecurity, with AI models not only detecting vulnerabilities, but also validating their exploitability and helping to develop effective remediation strategies. This integration into customer networks enables organizations to respond faster to threats, reduce uncertainty, and strengthen their overall security posture. In the future, similar initiatives are likely to become standard practice as AI technologies become increasingly integrated into the cybersecurity ecosystem.
Sources
- SiliconANGLE News – Palo Alto Networks to run OpenAI cyber models inside customer networks
- Yahoo Entertainment - Amazon (AMZN) Is Powering OpenAI's Cyber Models, So Why Is The Stock So Cheap?
