Arcjet Launches Runtime Security Controls for AI Agents
Summary
Arcjet introduced a product for monitoring AI-agent workflows, enforcing policies around actions such as tool and database calls, and preserving execution context for security reviews and compliance audits.
Key points
- Arcjet says its product can ingest agent activity through existing OpenTelemetry tooling without application code changes or another deployed agent.
- It connects activity across sessions and provides an inventory of agents and applications, including workflow actions, inputs, identity, and security decisions.
- Teams can define policies for prompt-injection detection, sensitive-data leakage and redaction, rate limits, quotas, and other controls.
- Policies can restrict actions such as email recipients, refund amounts, or web-fetch destinations; applications can stop an action, request human approval, or return an explanation.
- Arcjet says controls can be applied before and after calls to LLMs, tools, databases, and APIs.
- The product collects execution context and policy evaluations to support workflow reconstruction, security reviews, and compliance audits.
Article Details
- Event Type
- Product launch: Arcjet introduced agent runtime security for AI agents in production.
- Impact
- The product provides visibility into agent activity, deterministic policy enforcement before and after agent actions, and execution records to support security reviews and compliance audits. Policies can detect prompt injection, prevent or redact sensitive-information leaks, and control actions such as API calls, database operations, and refunds.