ChannelLife New Zealand - Industry insider news for technology resellers
New Zealand
Palo Alto launches continuous AI defence for clients

Palo Alto launches continuous AI defence for clients

Wed, 23rd Sep 2026 (Today)
Sean Mitchell
SEAN MITCHELL Publisher

Palo Alto Networks has launched Unit 42 Continuous Frontier AI Defence, an always-on version of its existing Frontier AI Defence offering.

The service is aimed at organisations using advanced AI models that want continuous testing of their security exposure rather than periodic reviews. It is designed to find, validate and remediate exposures and attack paths before attackers can exploit them.

Unit 42, Palo Alto Networks' threat intelligence and incident response division, introduced Frontier AI Defence earlier this year as a point-in-time exposure analysis service. The new version extends that approach with continuous offensive testing across changing IT environments.

The service uses a multi-model system that directs tasks to different AI models depending on the job. It includes access to Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6-Cyber, alongside open-weight models.

According to Palo Alto Networks, the service begins with a baseline scan of a customer's environment and continues testing as systems change. It is intended to assess first-party and third-party web applications, APIs, cloud infrastructure, source code repositories and network assets.

It also aims to verify whether weaknesses are practically exploitable rather than simply flagging theoretical risks. The service then prioritises fixes, provides code-level guidance and suggests virtual patches where needed.

Threat pressure

The launch comes as security vendors argue that AI is shrinking the time attackers need to move from identifying a weakness to exploiting it. Palo Alto Networks said threat actors are using AI to compress breach cycles from weeks to hours in some cases.

The company said it developed and tested the approach over six months of internal work and more than 100 customer engagements. It added that this was backed by a USD $17 million investment in research and development and methodology work.

During internal deployment, Palo Alto Networks said it found what it described as a year's worth of exposures in three weeks. In customer assessments, it said Frontier AI Exposure Analysis identified exposures in every customer reviewed, with 37% classified as high or critical severity.

It also said most exposures came from first-party applications, while more than two in three exposures in third-party applications had no known Common Vulnerabilities and Exposures listing. This points to a security challenge beyond known software flaws, particularly in the custom and connected systems many large organisations now run.

Philippa Cogswell, Vice President and Managing Partner, JAPAC, Palo Alto Networks Unit 42, commented on the regional context for the launch.

"As digital transformation and AI integration continue to drive organisational growth across the Asia Pacific & Japan region, cyber adversaries are leveraging advanced AI capabilities to compress attack timelines from days and hours to minutes. In this rapidly evolving threat landscape, traditional periodic and point-in-time security assessments are no longer sufficient to safeguard organisational assets and effectively manage cyber risk. By integrating leading Frontier AI models including Mythos and GPT-5.6 directly into Unit 42's Continuous Frontier AI Defence offering, we enable organisations in JAPAC to automatically discover, validate and remediate hidden risks and exposures before they can be exploited," said Cogswell.

Model access

The service reflects a broader shift in cybersecurity towards combining human-led testing with AI systems trained for specialist security work. Palo Alto Networks is positioning the offering around access to gated models from OpenAI and Anthropic, coupled with Unit 42's threat intelligence and offensive security teams.

Sam Rubin, Senior Vice President of Unit 42 at Palo Alto Networks, framed the issue as a mismatch between the speed of attackers and the pace of traditional defence.

"AI has created an asymmetric advantage for threat actors against organizations trying to defend at human speed. Modern cybersecurity requires machine-speed defense. We're combining Unit 42's offensive testing and threat intelligence expertise with industry-leading AI harnesses and exclusive access to gated capability models to give defenders back the upper hand," said Rubin.

OpenAI and Anthropic both backed the launch with statements underscoring how AI developers are seeking routes into operational security services through partnerships. The arrangement also suggests a growing commercial market for cyber products built around restricted-access models rather than only general-purpose AI tools.

"Frontier AI is compressing the time it takes to find and exploit vulnerabilities, and defenders need frontier AI capabilities within the tools, workflows and services they already trust. Through Daybreak and our work with Palo Alto Networks' Unit 42, we are pairing OpenAI GPT Cyber Models with deep security expertise, strong governance and human judgment to help organizations validate the attack paths that matter and move from discovery to remediation at machine speed," said McIntyre.

Anthropic focused on its model's ability to identify flaws that had previously gone undetected and on Unit 42's role in turning those findings into practical action.

"Claude Mythos found flaws that survived decades of human review, and more than ten thousand high-severity vulnerabilities across the software the world runs on. That kind of visibility is only useful if someone can act on it, and Palo Alto Networks is built for exactly that. Unit 42 takes what Claude Mythos finds, tests whether it can actually be exploited, maps what an attacker could reach from there, and gets the fix in front of the right team first," said Moore.

Palo Alto Networks said Unit 42 Continuous Frontier AI Defence is available worldwide on an annual subscription, with options that vary according to the models used.