Discussion

Semgrep previews Mandoline, a code-analysis tool built for AI security agents

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#12983

Semgrep has announced a research preview of Mandoline, a static-analysis tool designed for security agents and large language models. Its central promise is practical: help AI systems trace how data moves through code, across 50 programming languages.

Watch Desk analysis

What happened

Mandoline offers code slicing, dataflow analysis and code facts, according to Semgrep’s announcement. The company says the engine is significantly faster than previous static-analysis engines and is intended to help frontier AI models check software security properties efficiently.

It is a research preview, not evidence that the tool has already improved security outcomes in deployed software. Semgrep’s announcement describes its intended capabilities and performance, rather than independent test results.

Why it matters

Security agents need more than the ability to read code: they need ways to follow information through it and check whether it creates a vulnerability. Mandoline’s focus on dataflow and code slicing is aimed at that gap, while support across 50 languages gives it a potentially broad patch of software to work on.

If the speed and analysis claims hold up in practice, this could help AI systems examine code more usefully. That is a meaningful step beyond asking a model to spot trouble by staring at a repository and hoping for the best.

Our read

Mandoline is worth watching because it is built around a concrete technical job for AI security tools, not simply dressed up as an AI feature. But a research preview is an invitation to test the approach, not a security verdict. Developers should look for reproducible comparisons and examples of flaws the tool can actually help agents verify.

What to watch

  • When Semgrep makes the research preview available and what access it offers.
  • Whether the company publishes benchmarks supporting its speed claim.
  • How well Mandoline’s analysis works across the 50 languages it says it supports.
  • Whether security agents using it can demonstrate more reliable vulnerability checks.

Discussion spark: For AI security agents, what matters more: broad language coverage, or independently demonstrated accuracy on real vulnerabilities?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.