Adversa AI, an AI security pioneer with five years of offensive research, has launched the Adversa AI Coding Agent Security Platform for Developer Endpoints, generally available today.
Adversa AI has launched its Coding Agent Security Platform for Developer Endpoints, generally available today.
The platform connects what coding agents and agentic IDEs such as Claude Code and Codex are asked to do with what they actually do in real time.
It blocks the dangerous step before it executes, exposing attack chains and rogue agents whose steps look routine alone.
The platform covers observability and provenance via an agent behavior graph and semantic threat layer, with over 300 default policies across Security, Reliability and Privacy.
Context Guard detects attacks at a 0.1% false positive rate, outperforming Claude-as-a-judge, Azure Guardrails, AWS Bedrock and open-source guardrails on benchmarks from Microsoft, ETH Zurich and NVIDIA at 100x lower latency.
At a financial institution securing Codex, Claude Code and Copilot, blocking decisions took a median 0.2 ms, with correlation cutting noise to one confirmed incident per 2,000 raw signals.
In week one, the platform stopped an attack where an agent loaded a malicious skill, read internal API secrets and attempted to exfiltrate them.
"Coding agents deliver enterprise AI's quickest wins and biggest provable efficiency gains yet rank among the riskiest agentic AI systems according to AIRQ," said Alex Polyakov, co-founder and CTO of Adversa AI. "So we focused all our offensive experience on the immediate coding agent security challenge: control for security and risk teams, productivity and operational safety for engineering with minimal noise rate set as our slogan - cut the noise, deliver value."
The platform covers four essential components:
Observability and Provenance via Agent behavior graph and Semantic threat layer: In-memory engines evaluate each action against that graph and over 300 default policies covering every known coding agent attack technique, CVE, public incident and threat.
Agent Policy enforcement and Context Guardrails: An ensemble of in-memory engines making sub-millisecond blocking decisions, with Context Guard detecting attacks at a 0.1% false positive rate.
Universal agent runtime surface coverage: Via LLM proxy, MCP gateway, harness hooks, endpoint runtime and agent vendors' compliance APIs mapped into one universal agent behavior graph, following the AARM standard by CSA.
Agentic supply chain security and Shadow AI discovery: Detecting and managing all MCP servers, skills, hooks, plugins, instruction files, dependencies and generated artifacts.
About Adversa AI
Adversa AI lets enterprises ship AI without shipping risk. The company works with Fortune 500 companies, financial institutions, and AI scale-ups. Winner of 20+ prestigious awards including CSA Pitchapalooza, Fast Company Next Big Thing in Security, IDC AI Security Innovator, and recognized by Gartner, Wavestone and KuppingerCole. It leads the IEEE AI Cybersecurity, CoSAI Agentic AI Security workstream, is a core member of OWASP AIVSS, and contributes to CSA and NIST.
