Actual-time enforcement throughout the total agent improvement lifecycle — governing what brokers use, what brokers do, and securing the code they generate.
BOSTON, June 23, 2026 — Snyk, the AI safety firm, immediately introduced Evo Agentic Improvement Safety (ADS), extending safety protection to the AI workforce powering trendy software program improvement. Evo ADS secures how software program is constructed within the age of autonomous AI brokers — governing what brokers use, what they do, and what they generate — in actual time, contained in the agent workflow, earlier than dangers improve.
AI coding assistants have advanced into autonomous brokers — methods that invoke exterior instruments, take actions, and generate software program with minimal human oversight, linked to inside methods by MCP servers, plugins and third-party integrations. Current safety fashions have been constructed to scan code and artifacts, to not govern the methods creating that code, vet the instruments they use, or implement coverage on what an agent does at runtime. Knowledge from enterprise environments confirms the shift is already underway — and publicly documented assaults have already demonstrated working exploits by the agent toolchain itself, from malicious MCP servers to immediate injection embedded within the dependencies brokers devour.
Autonomous brokers have outpaced the safety fashions designed to control them
Anonymized telemetry from practically 9,700 developer environments reveals how far this shift has already progressed. Forty-three % of builders run two or extra AI coding environments concurrently, and greater than half have MCP servers put in. Probably the most instrumented environments had greater than 80 MCP servers working concurrently. These connections create reside entry to code repositories, browsers, inside instruments and manufacturing methods, with no safety controls between them.
The chance inside that provide chain could be very actual. One in 12 builders with MCP servers has a excessive or crucial discovering. A separate evaluation of early ADS enterprise design associate environments revealed the parallel menace in agent abilities: practically 1 in 4 builders has at the very least one talent put in, averaging 18 every, and greater than 1 in 10 abilities reference exterior dependencies or externally hosted directions. Current safety instruments scan code after it’s written; they haven’t any visibility into MCP configurations, abilities, or what brokers do at runtime.
Evo Agentic Improvement Safety: enforcement constructed into the agent execution loop
Till now, safety groups have confronted a false selection: block AI coding brokers completely and sacrifice the productiveness beneficial properties the enterprise is demanding, or permit them with no visibility into what they’re connecting to or what they’re doing. Evo ADS introduces a 3rd possibility: govern them. It introduces a steady management layer that operates contained in the agent workflow — not downstream from it — throughout three layers of the agentic improvement lifecycle:
-
Safe the agent provide chain: Discovers and assesses the MCP servers, abilities, and exterior instruments brokers pull in — surfacing immediate injection, malicious code patterns, and provide chain dangers earlier than brokers ever work together with them.
-
Govern agent habits: Screens and enforces real-time coverage on what brokers do whereas they function — blocking harmful actions earlier than they execute, and governing the methods brokers entry and the workflows they run.
-
Guarantee trusted output: Scans and fixes AI-generated vulnerabilities for the time being of creation — imposing safety at inception reasonably than in post-production evaluate.
One resolution. Three layers. Steady enforcement throughout each part of agentic improvement. Inside AI gateways can present routing and logging — however they can not decide whether or not an MCP server is malicious, whether or not a talent carries hostile directions, or whether or not generated code is definitely exploitable. That requires an unbiased enforcement layer working throughout real-world environments at scale.
“Ask a safety chief for an entire stock of the AI brokers, MCP servers, and abilities working throughout their developer machines — in most organizations, that stock would not exist,” stated Manoj Nair, Chief Expertise & Innovation Officer at Snyk. “That’s the hole Evo ADS closes. It discovers what is definitely put in, governs what brokers do whereas they run, and validates what they produce. The query is now not whether or not your group is utilizing AI brokers. It’s whether or not you have got a governance layer — and proper now, for many organizations, the reply isn’t any.”
For organizations already deploying AI coding brokers, the governance hole is rapid. Relay Community, whose engineering groups run GitHub Copilot, Codex, and Windsurf, and are transitioning to Claude Code as their major improvement assistant, embedded Snyk immediately into AI-assisted improvement workflows to implement safety as code is created.
“As we expanded our use of agentic improvement, it opened up a brand new assault floor,” stated Brendan Putek, director of DevOps, Relay Community. “We’re seeing provide chain assaults, malicious abilities and compromised MCP servers driving in on the agent’s personal toolchain, plus brokers taking actions with no guardrails between intent and execution. The blast radius is not bounded and we’re early within the curve. Working with Snyk, we landed on what I believe is the suitable structure: controls constructed immediately into the agent workflow that govern what an agent makes use of, executes, and generates.”
“Agentic improvement safety represents a elementary shift in how builders take into consideration code,” stated Oliver Neuberger, Managing Director, EMEA and UKI CMT cybersecurity observe lead, Accenture. “The potential for brokers to ship worth is big, however their affect calls for aware improvement and the suitable guardrails — so enterprises can deploy them securely and with confidence.”
Availability
Evo ADS shall be typically accessible June 29, timed to roll out whereas Snyk attends the AI Engineer World’s Fair, the place Snyk is the unique sponsor of the occasion’s first-ever safety observe.
The launch means the Snyk AI Safety Platform now governs AI throughout the total software program lifecycle, from brokers writing code to the fashions working in manufacturing to the purposes the brokers construct.
Evo AI-SPM delivers complete visibility and machine-speed governance of AI property. Evo Steady Offensive Safety (COS) simulates assaults to search out exploitable vulnerabilities earlier than adversaries do. Evo ADS secures the AI workforce itself — the brokers and instruments by which software program is now created. Collectively, the three options type the AI Safety Material – the unbiased validator that makes AI-generated code, AI brokers, and AI-native purposes reliable.
To study extra, learn our weblog on Evo Agentic Improvement Safety. To see Evo ADS in motion, register for our upcoming webinar, Securing AI-Pushed Software program Improvement within the Age of AI Brokers, on July 16.
About Snyk
Snyk, the AI safety firm, empowers the AI-driven enterprise to develop and safe its future, guaranteeing organizations can belief AI to innovate with out limits. The Snyk AI Safety Platform delivers the business’s AI Safety Material, weaving safety immediately into the stream of creation to safe GenAI code, AI-native purposes, and agentic methods. By delivering visibility, management, and autonomous protection safe at inception, Snyk permits over 4,800 international prospects to construct fearlessly within the AI period.