A Conversation With Prajakta Jagdale

Summary

Prajakta Jagdale, VP of Information Security at Palo Alto Networks, discusses how advanced AI models are changing offensive and defensive security. Her team’s early access to Anthropic’s Mythos Preview showed how capable frontier models can be at finding and exploiting vulnerabilities.

Key points

• Attackers may still have a slight advantage. Both attackers and defenders can use AI to discover vulnerabilities, but attackers can move directly to exploitation. Defenders must also triage findings, coordinate fixes, manage releases, and ensure patches reach every affected version.
• Organizational challenges often slow remediation. Ownership disputes, concerns about breaking applications, legacy systems, and lengthy approval processes can delay fixes. AI-generated proofs of concept can help demonstrate that a finding is real and harder to dismiss.
• AI can strengthen defensive workflows. It can help prioritize vulnerabilities by exploitability, pre-triage incoming reports, verify findings, and test patches. Jagdale argues that defenders need scalable processes—not just faster vulnerability discovery—and should address similar instances across systems at once.
• Security should be built into deployed systems. Jagdale supports secure base images, which her team said could reduce unpatched CVEs by about 80%. AI-assisted supply-chain pipelines can also map newly disclosed vulnerabilities to an organization’s assets and help identify patches or mitigating configuration changes.
• Layered protection remains important. She recommends balancing proactive security hygiene with detection and response. Examples include broad Cortex XDR coverage, firewalls, cloud protections, and controls that route software downloads through centralized artifact repositories.
• Automatic enforcement can prompt action. Palo Alto Networks automatically disables certain cloud resources that lack Cortex XDR and restricts internet-exposed resources with firewall rules. Enforcement deadlines can encourage engineering teams to prioritize remediation. The security team also tries to minimize disruption by scheduling changes during quieter periods.
• AI can be applied throughout development. Frontier models can help validate code-scanning findings by attempting to exploit them, identify patterns for higher-priority scanning rules, inspect commits and code graphs, and penetration-test builds after deployment. The company is still assessing how to integrate AI fully into CI/CD.
• Organizations need to prepare for agent-related risks. Jagdale says better visibility and observability are needed to distinguish legitimate from illegitimate agent activity. Canary techniques can help detect code or intellectual-property leaks, though their placement and integration require coordination.
• Product and security teams can reinforce each other. Jagdale cites security assessments and incident experience as inputs to products such as firewalls, Cortex XDR, and Prisma Browser.

Speakers and sources

Daniel, host of Unsupervised Learning, interviews Prajakta Jagdale, VP of Information Security at Palo Alto Networks. Unit 42 is mentioned as a source of threat research, and Cortex Cloud is cited as a Palo Alto Networks product resource.

Reviews, guides, or tutorials

None are presented. This is an interview and analysis discussion rather than a product review or step-by-step guide.