Endor Labs is a Leader in The Forrester Wave™ for Agentic Development Security

Endor Labs has been named a Leader in The Forrester Wave™: Agentic Development Security Tools, Q4 2026. We see the report as an early look at how security is changing as coding agents take on more of the agentic development lifecycle (ADLC).

Forrester gave Endor Labs the highest score possible in seven criteria:

• AI SCA
• AI application flaws
• AI triage
• AI prioritization
• AI coding toolchain
• Innovation
• Pricing flexibility and transparency

You can download a complimentary copy of the report here.

A new category for a different software development lifecycle

Agentic development security covers more ground than application security testing has in the past. Along with the code itself, it has to account for the coding agents that write it, the packages and tools those agents use, and the workflows they run in.

An AI coding agent can choose a dependency, install it, modify application code, and run commands before a change ever reaches a pull request. Each step introduces security decisions that used to sit largely with a human developer.

The agentic development lifecycle (ADLC) brings those decisions closer together. Security teams need to understand both what agents are doing and how their changes affect the application. That expands the job of application security. Reviewing generated code remains essential. So does governing the tools and permissions agents use, evaluating the packages they install, and giving them useful guidance while they work.

We see the emergence of agentic development security as a response to this architectural change. Organizations need security that participates throughout development and carries enough context to help resolve the risks it finds.

Security belongs where agents work

Organizations are adopting coding agents across editors, development environments, and cloud workflows. Security needs to fit those environments while giving teams a consistent way to express their requirements.

Endor Labs connects security capabilities to coding workflows through MCP, skills, CLI, and APIs. Our coding agent governance capabilities also use native hooks to apply policy to supported agent actions, while Package Firewall evaluates packages at installation time. Code is continuously reviewed and secured via plugins for every major coding agent, including Claude Code, Codex, Cursor, Devin, and many others.

Those controls serve different purposes. Governing an agent’s actions protects the development environment. Scanning code and dependencies helps assess the application it produces. Both matter as teams give agents more responsibility.

Our approach is to bring these capabilities together around application context, so security guidance can lead to a useful action and a result the team can review.

Our thesis: the harness beats the model

Writing code has become the least expensive part of agentic development. When researchers at Concordia University measured where tokens go across 30 agentic software development tasks, the automated code review stage consumed 59.4% of tokens on average, while writing the initial code consumed 8.6%. Most of the work, and most of the cost, now sits in verification.

Security review is part of that verification, and a frontier model working alone has clear limits. A model reasons well about the code it is shown, but it has no reliable way to decide which code matters or to follow data across files and dependencies. In our benchmark, frontier models examined less than 10% of the security-relevant code in a large enterprise Java codebase. In our Agent Security League, which extends the SusVibes benchmark developed at Carnegie Mellon University, the top agent as of October 2026 (Claude Code with Fable 5.1) wrote patches that passed functional tests 87.2% of the time, but only 37.4% also passed the security tests.

Our benchmarks show advancements now come from the harness more frequently than the model for coding tasks. That’s why we’ve invested in the security harness for agentic development. At the core is our patented code context graph, a continuously updated map of how code, dependencies, and data connect, which points agents at the paths that can actually lead to a vulnerability and gives them evidence to check their work. 

Our AI SAST benchmark shows what that adds. Across eight real-world projects in six languages, with ground truth verified by our security research team, Endor Labs AI SAST:

• Found 2.6x more real vulnerabilities than Claude Opus 4.7, including 63 that no other tool or model in the test caught
• Covered 1.8x more weakness types (CWEs) than the next-highest model or tool
• Concentrated its unique findings in broken access control, insecure direct object references (IDORs), authentication failures, and business logic flaws

The harness also makes the model more efficient. Across 34 security prompts, the same model used 12x fewer tokens with Endor Labs evidence than it did unaided (6.6 million versus 79.5 million) and finished in about a third of the time.

That efficiency also shapes how we price. Endor Labs is priced per contributing developer, and new AI capabilities are included without separate token-based billing, so teams can expand how they use security agents without watching a token meter. Forrester gave Endor Labs the highest score possible in the pricing flexibility and transparency criterion.

Building the next chapter with our customers

We’re grateful to the customers who challenge us to make security more useful in the way they build software. Their work keeps our focus concrete: help engineering spend less time on security tasks, and helping security teams spend more time on strategic work by giving them evidence to act.

We’re proud to be recognized as a Leader in this Wave. Our ambition is to help organizations build an ADLC in which security decisions are informed, policies are enforceable, and changes can be checked before they ship.

You can download a complimentary copy of the report here.

Forrester does not endorse any company, product, brand, or service included in its research publications and does not advise any person to select the products or services of any company or brand based on the ratings included in such publications. Information is based on the best available resources. Opinions reflect judgment at the time and are subject to change. This report is part of a broader collection of Forrester resources, including interactive models, frameworks, tools, data, and access to analyst guidance. For more information, read about Forrester’s objectivity here.

‍