Arcjet Launches Runtime Security for Autonomous AI Agents

The platform introduces a framework focused on observability, enforcement, and audit capabilities. By integrating with existing OpenTelemetry tooling and native agent frameworks—including those from OpenAI, Microsoft, and LangChain—Arcjet allows security teams to monitor agent activity without requiring extensive code changes. This visibility tracks sequences of actions across sessions, effectively turning a string of isolated events into a coherent workflow audit trail.

Security teams can now implement deterministic policies to intercept and validate actions before they reach APIs, databases, or LLMs. These controls mitigate risks like prompt injection or unauthorized data exposure by enforcing rules on sensitive operations, such as refund amounts or email attachments. According to CEO David Mytton, the platform is designed to connect disparate steps that might appear harmless in isolation but pose risks when executed in sequence. The system uses the Open Policy Agent framework to maintain versioned, immutable rules, ensuring that security teams can trigger human approvals or halt operations immediately when an agent deviates from defined safety parameters.