AI Is Now Hunting Weaknesses in Corporate Networks
Palo Alto Deploys GPT-5.6-Cyber and Claude Mythos 5
Frontier artificial-intelligence models are now being put to work continuously attacking corporate computer systems — with permission.
Cybersecurity company Palo Alto Networks has launched Unit 42 Continuous Frontier AI Defense, a service that uses advanced AI models to repeatedly search customer systems for vulnerabilities and determine whether those weaknesses could actually be exploited.
The service uses OpenAI's GPT-5.6-Cyber, Anthropic's Claude Mythos 5 and open-weight AI models through a proprietary system that routes different security tasks to whichever model is considered best suited to them.
Rather than relying only on periodic penetration tests, the system performs continuous testing as an organisation's environment changes. It can examine web applications, APIs, cloud infrastructure, source-code repositories and network assets.
The models can attempt to validate vulnerabilities, identify combinations of weaknesses that form viable attack paths and recommend fixes, including code-level changes and virtual patches. Palo Alto says Unit 42 offensive-security specialists remain involved in directing and validating the work.
The approach reflects a rapidly developing contest in cybersecurity as attackers increasingly use AI to accelerate vulnerability discovery and exploitation.
Palo Alto says that in one recent investigation an attacker used more than 50 recognised attack techniques and compressed activity that could otherwise have taken a skilled red team weeks into less than 10 hours.
The company also says its earlier Frontier AI Exposure Analysis found security exposures in every customer environment assessed. Of the exposures identified, 37 percent were rated high or critical in severity. Those figures come from Palo Alto's own customer assessments rather than independent testing.
OpenAI restricts access to GPT-5.6-Cyber, describing it as a specialised model for approved defenders conducting authorised vulnerability research, exploit validation and security testing.
Unit 42 Continuous Frontier AI Defense is available worldwide through annual subscriptions, with different options using combinations of OpenAI, Anthropic and open-weight models.
