Adversa AI Coding Agent Security Platform

Adversa AI Ltd. offers a Coding Agent Security Platform that acts as a runtime control layer for enterprises deploying AI coding agents such as Claude Code, Copilot, Cursor, and Codex. It captures LLM calls, tool executions, and endpoint actions, connects them into chains, and blocks dangerous ones before they complete. The vendor states the platform is in early access with first production deployments live.

What It Is

The platform is an AI detection and response product built for coding agents. Rather than inspecting prompts alone, it builds an agent behavior graph correlating model conversations, tool calls, and endpoint actions, and scores risk across the linked chain. Detections ship with the chain, the rule that fired, and the evidence behind it.

How the See, Understand, Stop workflow works

• See: Watches model traffic (including tool and MCP calls), the agent harness, the developer's machine, and vendors' compliance APIs.
• Understand: Correlates events into chains, so a lone suspicious step is context rather than an alert.
• Stop: Blocks a chain when its risk score crosses a configured threshold; enforcement can be scoped per policy, environment, or team.

Adversa says over 300 policies ship by default, and teams can describe policies in plain English, which the platform converts into enforceable policies that are validated, simulated, versioned, and approved before going live.

Deployment model

The vendor describes three components: an agent security gateway on agent traffic (inline or reading from an existing gateway), endpoint sensors on developer machines that watch only agent actions, and harness hooks plus provider integrations. It runs on customer infrastructure or in the cloud. Pilots begin in observe mode, with blocking enabled when the customer is ready. The vendor states the design fails open and that SIEMs can ingest the event stream.

Compliance and roles

The vendor maps its action records and pre-execution blocking to requirements such as EU AI Act Articles 12 and 14, DORA, and SR 11-7, and to frameworks like NIST AI RMF, ISO 42001, and SOC 2. It targets AppSec teams, CISOs, engineering leaders, and risk and compliance teams.