What happened
On Oct 29, Adversa AI launched generally available its Coding Agent Security Platform, which links what coding agents (Claude Code, Codex, Copilot) are told to do with what they actually do in real time, building an agent behavior graph and blocking dangerous steps before execution with sub-millisecond in-memory engines over 300+ policies.
Why it matters
Coding agents are among the riskiest agentic systems, and this provides real-time prevention of multi-step attack chains (0.1% false positive rate, 100x lower latency than judge/guardrail baselines), plus universal runtime coverage via LLM proxy, MCP gateway and harness hooks, following the CSA AARM agentic runtime standard.
Applicability
AppSec/DevSecOps teams adopting coding agents should evaluate it for agent provenance, context guardrails and agentic supply-chain discovery; a deployed bank case shows median 0.2ms decisions.