Sysdig launches AI Defense to monitor and govern autonomous AI agents

From chat to action

Artificial intelligence is shifting from a conversational tool to an active participant in business workflows. While previous models were primarily chat machines, new agentic AI systems are designed to perform specific tasks autonomously. These agents can ship code, modify infrastructure, and reach into production systems at speeds that make manual review impossible.

This shift introduces significant risks because agents lack a sense of morality and can act faster than human teams can control. As agents are designed with more autonomy and reasoning capabilities to maximize economic benefits, the potential for them to drift into errors increases. Without proper oversight, these tools could quickly transform from business assets into operational disasters.

Sysdig has entered this space with the launch of Sysdig AI Defense. The company aims to provide runtime security and governance for these agents by monitoring their behavior in real time. This approach ensures that as AI begins to do more for companies, it remains within the boundaries of intended use.

The kernel as a source of truth

Monitoring an AI agent is complex because an agent can assume new powers very rapidly to complete a task. While every action an agent takes might be logged, the underlying processes are often only visible to the operating system kernel. These hidden processes could be invisibly changing how an agent functions during its execution.

Sysdig notes that an agent deployed in the morning might not be the same one running in the afternoon. Because only the kernel shows what actually ran, the software must be watched in real time to ensure it stays on track. This provides a layer of truth that goes beyond what the agent reports about its own actions.

The system is designed to identify discrepancies between an agent's reported intent and its actual behavior. If an agent misreports a tool call, the system can detect the mismatch between the report and the system interaction. This detection allows for immediate intervention when an agent steps out of line.

Real-time enforcement and logs

For safety to be effective, any blocking response must happen in real time as the action occurs. Sysdig AI Defense is built to operate in line with the action, stopping it before it can complete. This allows organizations to prevent unauthorized changes to their systems without delaying the entire workflow.

The product offers several specific benefits for managing these autonomous systems. It creates a live inventory of agents, harnesses, models, MCP servers, skills, and plugins based on runtime telemetry. Furthermore, it ensures that every agent can be traced back to a specific, accountable human user.

Organizations can use predefined policy templates mapped to established frameworks like the OWASP Top 10 for LLM Applications. These policies can start in an alert mode to monitor behavior before being moved into an enforcement mode. This allows teams to gradually tighten controls as they become more comfortable with the agent's performance.

Visualizing the action graph

One of the key features of the new tool is a live action graph that shows each step an agent takes in order. This graph pairs what the agent said it was doing with what it actually did and what it was able to reach. It also identifies whose authority the agent was acting under at any given moment.

These interactions are recorded as they happen to create audit-quality, immutable records. These logs show which policies were applied, when they were triggered, and who signed off on the permissions. This level of detail is necessary for compliance and for understanding the history of autonomous decisions.

As securityweek.com reported, this visibility is crucial because manual governance cannot keep up with the volume of decisions. The goal is to provide a clear map of the agent's journey through the system. This ensures that even as agents become more complex, their actions remain transparent to the humans who deployed them.

The scale of autonomous AI

The need for automated governance is driven by the rapid growth of AI deployment in the enterprise space. The IBM Institute for Business Value reported that enterprises expect to deploy an average of 1,661 AI agents by 2027. This represents a 38% increase from current levels, creating a massive management challenge for tech leaders.

At this projected rate, companies will be tasked with managing hundreds of thousands of autonomous decisions every single day. Manual supervision is no longer a viable option for this volume of activity. Automated systems are required to provide the necessary oversight to keep these agents secure.

Sysdig AI Defense addresses this by providing a foundation for automated runtime security. By focusing on the truth of what actually runs on the system, it helps organizations manage the complexity of agentic AI. This allows businesses to harness the economic benefits of AI while mitigating the risks of autonomous errors.