Snyk, the AI security company, today announced Evo Agentic AppSec, a team of AI security agents that finds, fixes and blocks vulnerabilities at the speed agents write code, with every result verified independently of the agent that produced it. Agentic development has outpaced the AppSec model built for human-written code. Evo Agentic AppSec is built to secure agent-written code at inception and to provide ongoing coverage of existing, human-written code and it runs on the deterministic engines more than 4,800 customers already rely on.

This puts backlog zero within reach for enterprises for the first time ever.

“Snyk started with a simple thesis. Every defect is cheaper to fix the closer it sits to the moment of creation. Agents write the code now, so security has to shift in, inside the loop where code is created. And the test is not whether a fix was generated. The test is whether it ships. A model grading its own homework is not a security program. It is a conflict of interest with a dashboard,” said Manoj Nair, Snyk CTO and chief innovation officer.

Pointing general-purpose LLMs at the problem won’t close the gap. Snyk’s latest VulnBench report shows that 49.7% of the time, LLM security reviews report a finding that disappears on the next run, meaning they cannot be the deterministic engines that serve as the foundation of issue discovery. Today, with AI-driven attacks, fixing issues is the bottleneck. In separate research, VulnBump, Snyk demonstrates that LLMs alone produce a mergeable fix only 59.6% of the time, a rate no engineering team would accept from even its most junior developer. In the same benchmark, held to the same mergeable fix standard, Evo Remediation Agent, designed to autonomously tackle security backlogs, raised that rate to 90%.

Why now

New critical and high-severity security issues detected across Snyk customers grew 10x in the past year, and analysis of Snyk’s product data showed roughly 8 new vulnerabilities introduced for every one remediated. The gap keeps widening as agentic development spreads.

Attackers have moved at the same pace. Snyk researchers have already responded to 18 high-profile malware-focused attacks in 2026, and project a 500%+ annual increase in malicious packages published to open-source registries. One threat actor breached more than 600 FortiGate firewalls across 55 countries in five weeks using off-the-shelf AI models. Snyk’s analysis of nearly 10,000 developer environments found that one in 12 developers running MCP servers had a high or critical security finding, and its audit of nearly 4,000 agent skills found at least one security flaw in 36% of them.

“A pull request scan was built for human speed and human volume,” Nair said. “It still matters. It was never going to be enough on its own once code was generated continuously and attacks were automated against it.”

“The backlog you have been managing for years is not a cleanup project anymore. It is an attack surface,” Snyk CEO Ken MacAskill said. “Backlog zero used to be a fantasy, but in the era of agentic development and attacks it must be the standard.”

What Snyk is shipping

Evo Agentic AppSec works the application security loop at machine speed, with humans on the loop: understand the application, model its threats, find the vulnerabilities that matter, decide what is worth fixing, and ship validated, mergeable fixes across both new code and the backlog already in production. The agents do not replace Snyk’s deterministic engines. They operate them, calling Snyk Code, Snyk Open Source and Snyk’s other engines as tools. Evo Agentic AppSec ships as a team of purpose-built agents:

Evo Code Scanning Agent pairs frontier AI reasoning with Snyk’s existing static analysis to catch what pattern-matching structurally cannot: broken access control and authentication logic flaws, concentrated in exactly the places AI-generated code is weakest.

Evo Dependency Firewall blocks malicious and out-of-policy packages at install time, before they ever reach a developer machine, let alone the codebase. That includes slopsquatting, where attackers register the package names AI agents hallucinate and wait for the next agent to install the payload.

Evo Agentic Security Review delegates a full security review of a repository to an agent, builds an application model and threat model from code and context, assesses whether controls are present and sufficient against credible threat scenarios, and produces evidence-backed findings with concrete remediation guidance, independent of scanner coverage.

Evo Remediation Agent runs the full fix loop autonomously and at scale, across both newly written code and the backlog, starting with Snyk Open Source issues at launch. A fix that might break the build is queued for approval, not merged, so Remediation Agent is built to deliver mergeable fixes.

Evo Secrets Detection catches leaked credentials across both traditional and agentic development, from the IDE through CI.

Each of these agents draws context from an intelligence layer with more than a decade of fix-outcome data, and runs reachability, exploitability and breakability analysis, exposed as a resource any agent can call.

Snyk handles prevention and remediation differently. Prevention inside the coding agent needs no human sign-off: an agent already authorized to write code should not need to ask before fixing a vulnerability it just introduced. Remediation touches code that may already be running and that someone else is accountable for, so it has to earn that person’s trust before it ships.

One loop, end to end

Evo Agentic AppSec completes the Snyk AI Security Platform as one loop. It covers detection, remediation and the supply chain. Snyk Studio, adopted by more than 1,600 customers, offers real-time AI guidance and intelligent, conversational remediation in the command line. Evo Agentic Development Security (ADS), previously released, prevents vulnerabilities inside the coding agent before code is emitted. Evo Continuous Offensive Security, generally available since August, acts as the independent judge, attacking the running application the way an adversary would to prove fixes hold.

None of the agents grade their own work. “Agents do the work. Something independent proves it holds. Humans own the bar,” Nair said. “The generator cannot be the validator.”

The roles change, not the headcount

Evo Agentic AppSec is built to run without a proportional increase in AppSec staff as AI development scales. Developers move from author to approver of the fixes agents propose. AppSec leads move from triaging tickets to working on the loop as AI Security Engineers: setting the bar the agents operate against, governing the agents and auditing what they decide. Teams with no dedicated AppSec function get a working program for the first time.

Availability

Evo Code Scanning Agent, Evo Dependency Firewall and Evo Agentic Security Review are available now to select Snyk customers in experimental preview. Evo Remediation Agent is available now in open preview. Secrets Detection, Evo ADS and Evo Continuous Offensive Security are generally available. Organizations can talk to their Snyk account representative for access, or visit snyk.io to learn more.

About Snyk

Snyk, the AI security company, empowers the AI-driven enterprise to develop and secure its future, ensuring organizations can trust AI to innovate without limits. The Snyk AI Security Platform secures software at build speed, weaving protection directly into the creation flow to secure GenAI code, AI-native applications, and agentic systems. By delivering visibility, control, remediation and autonomous defense securely at inception, Snyk enables more than 4,800 global customers to build fearlessly in the AI era.

Media gallery

About The Author
Business Wire