It is claimed to have been verified in a production environment that the Agentic Development Security solution for Adversa can prevent encoding proxy attack chains within 0.2 milliseconds.

Tel Aviv-Yafo, Israel, October 29, 2026 / PRNewswire / -- Adversa AI, a security pioneer with five years of experience in offensive research, has today released Adversa AI Coding Agent Security Platform for developer terminals.

Adversa AI indicates that this platform is now officially available at https :// adversa.ai / agent-security-platform. It is capable of real-time linking encoding proxies with the tasks required to be executed by agentic and IDE (such as Claude Code and Codex) to their actual behaviors. This linkage can expose attack chains and proxy behaviors that appear normal but are actually abnormal, and it can prevent such actions before they are carried out in dangerous steps.

The platform encompasses four core components of Agentic Development Security.

• Observability and traceability are achieved through proxy behavior graphs and semantic threat layers. The memory engine compares each action with this graph as well as more than 300 default policies, which cover categories such as security, reliability, and privacy, and involve known encoding proxy attack techniques, CVE, public events, and threats.
• Real-time protection enables the execution of proxy policies and context-based safeguards. A system composed of multiple memory engines can make blocking decisions in sub-millisecond times. Adversa indicates that its Context Guard can control the false positive rate at 0.1%. In benchmark tests conducted by Microsoft, ETH Zurich University of Technology, and NVIDIA, it has a 100-times lower latency compared to Claude-as-a-judge, Azure Guardrails, AWS Bedrock, as well as open-source safeguard solutions.
• Through LLM proxies, MCP gateways, harness, hooks, terminal runtime, and the compliance requirements of proxy vendors API, the general proxy runtime surface is overlaid and mapped to a unified proxy behavior diagram. This platform adheres to the AARM proxy runtime security standards specified by CSA.
• By detecting and managing all MCP servers, skills, hooks, plugins, command files, dependencies, and generated artifacts, proxy supply chain security is ensured, as well as Shadow AI detection is achieved.

According to Adversa AI, in a financial institution that deployed this platform to protect Codex, Claude Code, and Copilot, the median latency of blocking decisions was 0.2 milliseconds, and no changes in developer experience were reported. From day one, the AppSec, product security, and DevSecOps teams were able to monitor each proxy action, and correlation analysis reduced noise to only 1 confirmed event out of every 2,000 raw signals. Within the first week, the platform prevented one attack: a proxy loaded malicious code, followed its hidden instructions, read an internal API key, and attempted to transmit it externally. Other security tools failed to identify this chain of events; however, the platform intercepted the data transmission request.

Adversa AI, Co-Founder and Chief Technology Officer, and Alex Polyakov stated: "Encoding proxies bring the fastest implementation results and the most verifiable efficiency improvements to AI. However, according to AIRQ, they also belong to one of the highest-risk agentic systems. Therefore, we are dedicating all our expertise to addressing the current security challenges associated with encoding proxies: providing control for the security and risk teams, enhancing productivity and operational security for the engineering teams, while minimizing noise levels – our slogan is 'Reduce noise, deliver value.'"

The platform's defense capabilities are based on a series of entirely new AI attack categories that the team has publicly identified, including Reasoning Leakage, PromisqRoute, Cryptographic Context Injection, SymJack, and TrustFall. These have now become part of the industry's threat landscape.

About Adversa AI

Adversa AI helps enterprises deploy AI without taking on risks. The company serves Fortune 500 companies, financial institutions, and AI growth-oriented firms. It has won over 20 awards, including CSA Pitchapalooza, Fast Company Next Big Thing, in, Security, and IDC AI Security Innovator, and has received recognition from Gartner, Wavestone, and KuppingerCole. The company's leaders IEEE AI Cybersecurity and CoSAI Agentic AI Security are core members of the OWASP AIVSS workflow and have contributed to research in CSA and NIST; their research has appeared in publications such as The Wall Street Journal, Wired, Bloomberg, and TechCrunch. Learn more.

Media contact: Adversa AI PR +97504794776 [ email protected ]