Palo Alto Networks launches always-on AI security testing built on Claude Mythos and GPT-5.6-Cyber

Palo Alto Networks Launches AI Security Testing with Claude Mythos (discover more) and GPT-5.6-Cyber

Palo Alto Networks’ Unit 42 is introducing Continuous Frontier AI Defense, a subscription service utilizing restricted access models from Anthropic and OpenAI to continuously scan customers' systems for vulnerabilities. This proactive approach aims to counter the growing threat of AI-powered attacks, as demonstrated by attackers leveraging similar AI technologies against corporate networks.

Michael Moore, Cybersecurity Lead at Anthropic, highlighted the capabilities of their model: "Claude Mythos discovered flaws that stood the test of time, identifying over 10,000 high-severity vulnerabilities across widely used software."

The service utilizes Unit 42's proprietary software to assign tasks to models, optimizing results and costs. Sam Rubin, Senior Vice President of Unit 42, emphasized the need for machine-speed defense in modern cybersecurity: "AI has tipped the balance, giving threat actors an advantage over organizations trying to defend manually."

Key features of Continuous Frontier AI Defense:

  • Initial Comprehensive Scan: Scans a customer's entire estate.
  • Continuous Monitoring: Continuously targets web apps, APIs, cloud infrastructure, code repositories, and networks as they evolve.
  • Exploitable Vulnerability Validation: Proves exploitability for each identified weakness.
  • Threat Modeling: Maps potential next steps for attackers.
  • Remediation Suggestions: Offers code fixes or temporary "virtual" patches.

Based on six months of development and 100+ customer engagements, Palo Alto Networks claims the service can surface as many vulnerabilities in three weeks as their traditional methods uncover in a year. They also assert that AI cuts the time from flaw discovery to exploitation from weeks to hours.

Anthropic's restricted access model, Claude Mythos, has been instrumental in finding over 10,000 high-severity vulnerabilities. Access was briefly suspended in June due to US export controls but restored on July 1st.

Pricing is subscription-based and varies based on the mix of models used and customer needs. The service is available globally.

Leave a Reply