Palo Alto’s Unit 42 turns Mythos 5 and GPT-5.6-Cyber into a continuous defensive harness
Sam Rubin, 22 September: always-on exposure discovery, exploitability checks and remediation. Internal: a year’s pentest work in three weeks; 3.2× more high/critical findings; 51 percent faster mean time to remediate. Two-thirds of validated third-party exposures had no CVE.
Short1 minIn-depth
Maya Chen1 min read
- Unit 42
- Palo Alto
- Mythos
- GPT-5.6-Cyber
- Security

Photo: Palo Alto NetworksSource page
Unit 42 on Tuesday launched Continuous Frontier AI Defense, an always-on service that uses Anthropic’s Claude Mythos models, OpenAI’s GPT-5.6-Cyber, and open-weight cyber models behind a proprietary multi-model harness. Sam Rubin’s blog post is the source. The pitch is discover, prove exploitability, then push a fix — applications, identities, cloud, network — as the estate changes. A single model, Unit 42’s own evals say, catches no more than 40 percent of vulnerabilities in a complex environment; Mythos 5 and GPT-5.6-Cyber overlap on less than 10 percent of the exposures they find. The harness routes jobs. Zero-data-retention, the company says, so customer code is not kept or used to train public models. The April Frontier AI Defense product was point-in-time. This one is a subscription.
Internal numbers, from Palo Alto’s own deployment: continuous Mythos scanning did more than a year of traditional pentest work in three weeks, found 3.2 times more high and critical vulnerabilities per product, and cut mean time to remediate 51 percent. Across more than 100 customer jobs, two of three validated exposures in third-party applications had no CVE; 37 percent of findings were high or critical. One financial-sector chain — a payment link that skipped re-auth, a skipped OTP, a session-routing flaw — was account takeover with no victim click, Rubin wrote. Those are vendor figures from the vendor’s customers. The service is available worldwide on an annual contract. It is a security-desk brief, not a second write-up of Opus 5.5’s cyber fallback.


