Palo Alto Networks, the US cybersecurity company, is rolling out an always-on service that uses autonomous AI agents to probe customers' systems for weaknesses.
The company describes it as an agentic offensive security service, meaning AI agents act like attackers, working through tasks on their own rather than waiting for human instructions.
The agents find exposures, confirm whether they can actually be exploited and flag them, so security teams can prioritise fixes before real attackers strike.
The service combines restricted-access frontier models, including Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6-Cyber, with open-weight models, whose underlying code is publicly available.
A proprietary system then picks the best model for each scan, Palo Alto said.
Unit 42, Palo Alto's threat intelligence and security consulting arm, will run continuous tests across web applications, cloud infrastructure, source-code repositories and APIs, the connections that let software systems talk to each other.
The service is sold as an annual subscription.
It extends Palo Alto's Frontier AI work, which earlier this year offered one-off assessments of customers' exposure, into continuous, agent-driven testing.
Palo Alto said the aim is to shrink the gap between finding a weakness and fixing it by automating offensive testing at scale.
The company said the service responds to a threat landscape in which hackers are increasingly using AI.
The service's name is missing from your copy ("rolling out, an always-on service"), so I've written around it; worth dropping it back into the first paragraph.
