Company news
Adversa AI Launches Coding Agent Security Platform to Block Cyberattacks on AI Agents in Real Time
Published 30 October 2026
WHAT HAPPENED
Adversa AI launched the Coding Agent Security Platform for Developer Endpoints on October 29, 2026. The platform is generally available and blocks cyberattacks on coding agents such as Claude Code, Codex, and Copilot in real time, with blocking decisions occurring in under 0.2 milliseconds. The platform was tested in production at a financial institution where it stopped an attack in week one involving a malicious skill attempting to exfiltrate internal API secrets.
WHO IS INVOLVED
Adversa AI, an AI security company based in Tel Aviv with five years of offensive research, built and launched the platform. The company works with Fortune 500 companies, financial institutions, and AI scale-ups. Alex Polyakov, co-founder and CTO, leads the effort and emphasized the platform's focus on control for security teams and productivity for engineering with minimal false positives.
HOW IT WORKS
The platform connects what coding agents are asked to do with what they actually do, exposing attack chains through an agent behavior graph and semantic threat layer. It enforces policies and context guardrails using in-memory engines that evaluate each action against over 300 default policies covering security, reliability, and privacy. The platform detects attacks across multiple agent runtime surfaces including LLM proxies, MCP gateways, and endpoint runtimes, and identifies shadow AI through supply chain security scanning of MCP servers, skills, and dependencies.
WHY IT MATTERS
Coding agents rank among the riskiest agentic AI systems yet deliver enterprise AI's fastest wins and biggest efficiency gains. Existing security tools missed the attack that Adversa's platform stopped in the financial institution deployment, demonstrating a gap in current defenses. The platform achieves a 0.1 percent false positive rate, outperforming alternatives including Claude-as-a-judge, Azure Guardrails, and AWS Bedrock guardrails while operating at 100 times lower latency.
THE PLATFORM
The Coding Agent Security Platform covers four essential components: observability and provenance via agent behavior graphs, policy enforcement and context guardrails for real-time prevention, universal runtime surface coverage across vendors via standardized formats, and agentic supply chain security with shadow AI discovery. The platform follows the AARM standard by the Cloud Security Alliance for agentic runtime security and integrates with multiple coding agents and development IDEs through a single universal agent behavior graph.
- Type
- Product launch
- Category
- AI Security
- Source
- adversa.ai ↗
