Practical DevSecOps launches agentic AI security certification

Practical DevSecOps has launched CAASE, a hands-on certification for defending AI agents after a survey found nearly 9 in 10 organizations reported or suspected an AI agent security incident in the past year. The program targets a fast-growing skills gap as agents move into production and take actions on real systems with real credentials.

Why it matters: - AI agents can plan, remember, call tools, run code and hand tasks to other agents, which creates attack paths that traditional security models were not built to cover. - Nearly nine in 10 organizations confirmed or suspected an AI agent security incident in the past year, according to Gravitee's State of AI Agent Security 2026 Report. - The reported incident rate was 92.7% in healthcare. - Infosec. QA's 2026 salary data shows specialists with agentic red-teaming or multi-agent security skills earn 20–30% more than standard AI security pay.

What happened: - Practical DevSecOps launched the Certified Agentic AI Security Expert, or CAASE, on Oct. 1, 2026. - The certification is designed to teach security professionals how to attack and defend AI agents end to end. - Pre-booking is open now. - CAASE is available at Practical DevSecOps.

The details: - The course runs across seven chapters. - Learners practice hijacking an agent's goal mid-task and detecting the attack. - Training covers poisoning long-term memory and retrieval-augmented generation sources, then hardening against those attacks. - Participants learn to escape a coding agent's workspace, spoof agent-to-agent messages and exploit confused-deputy chains. - The curriculum includes container sandboxing, scoped delegation tokens and skill vetting as defenses. - The course closes with threat modeling in MITRE ATLAS, AI agent bills of materials, and mapping controls to the NIST AI RMF and the EU AI Act. - Training also covers security for the A2A, ACP and UCP agent protocols. - The program is about 70% hands-on. - Learners receive 60 days of browser-based lab access. - The labs include more than 30 guided exercises. - Instructor support is available 24/7. - Certification requires a six-hour practical exam with five challenges. - A passing score is 80% or higher. - The certification is aimed at AI and agentic security architects, application security engineers, information security engineers, applied AI security engineers and penetration testers. - Basic Linux skills are required. - Scripting experience helps. - Practical DevSecOps recommends the Certified AI Security Professional, or CAISP, course first. - Organizations can train teams through the enterprise program. - Certification bundles save a minimum of USD 270.

Between the lines: - The launch reflects a shift from general AI security awareness to hands-on defense of autonomous systems that can take real actions inside production environments. - Practical DevSecOps is positioning CAASE as a skills-based answer to agent-specific threats such as memory poisoning, message spoofing and sandbox escape. - The focus on standards and frameworks suggests buyers will want training that maps directly to governance and compliance work, not just red-team tactics.

What's next: - Practical DevSecOps is now taking pre-bookings and selling team training through the enterprise program. - The company will likely use the certification to expand its agentic AI security training footprint as more organizations put AI agents into production.

The bottom line: - AI agents are creating a new security problem fast, and Practical DevSecOps is betting hands-on certification will help close the talent gap before incidents spread further.