Sysdig just launched Secure AI, aimed squarely at a problem that’s been getting worse fast: the gap between when a vulnerability gets disclosed and when someone exploits it keeps shrinking, and cloud security teams are struggling to keep pace working by hand.
Sysdig built Secure AI on top of its existing cloud-native protection platform, and the product works three ways. Users can set goals for AI agents to pursue independently, connect through coding assistants like Claude, Cursor, or Codex, or use a generative AI assistant that explains risks in plain language and suggests fixes. Sysdig says staff still make the final call on anything critical, but automation handles routine investigation and response, and the system logs every action for review.
Much of the case Sysdig is making rests on its own research. The company says its researchers watched an AI agent move from a single vulnerability to a full internal database breach in under an hour, and it points to JADEPUFFER, which it calls the first ransomware attack an AI agent ran entirely from start to finish. Whether that framing holds up under scrutiny is worth watching, but the underlying pressure looks real regardless: a survey from the Information Systems Security Association and Omdia found 83% of organizations are already adopting or planning to adopt AI for security work.
What separates Secure AI from other AI security tools on the market is its reliance on runtime data, essentially what’s actually running in a cloud environment at a given moment, instead of a long list of theoretical vulnerabilities that an attacker may never actually exploit. Sysdig combines that data with its own threat research and the open-source Falco detection engine, then tunes the system over time to learn how each organization operates.
The company backed the launch with specific numbers, too, claiming customers can run more than 10 times as many investigations as human analysts alone, at 88% lower cost. As one example, Sysdig described a vulnerability check that normally takes three analysts 45 minutes each, compared to one analyst finishing the same task in under 15 minutes using Secure AI.
Sysdig founder and CTO Loris Degioanni argued that traditional monitoring tools work like cameras, recording break-ins for people to review afterward, while attackers using AI now move faster than that model can react to.
