Home / Blog / Every Week Someone Asks Me When AI…
AI and LLM Security
Every Week Someone Asks Me When AI Will Replace Pentesters
Datadog released a scanner that sends your code to an LLM instead of matching patterns. After 20 years in this field: the engine was never the problem. What they cannot see is.
This piece is my read on research published by Datadog Security Labs. The findings are theirs; the argument about what they mean is mine. Their original write-up is worth reading in full.
Every week somebody asks me when AI is going to replace penetration testers. So here we are - the team at Datadog have just released a tool that tries to do exactly that.
The tool is called SAIST, Static AI Security Testing. Broadly? It is SAST without rules. Instead of an engine looking for patterns in code, the tool sends your code to an LLM - Claude, GPT or Gemini - using your own API key, and the AI looks for weaknesses. Results come back as SARIF, the whole project is indexed so there is context between files, and it currently supports Java, Python and Go.
Sounds like the future? Honestly, yes. But not the way you think.
A few things worth knowing before getting excited
- The tool is still in preview. Which means Datadog themselves do not yet trust it with their eyes closed.
- You cannot run it without a Datadog account, because the detection rules live in their cloud. An “open source” tool that does not work without the vendor is a flexible use of the term.
- Your code is sent to an external LLM. Stop for a second and think about what that means for the most sensitive code in your organisation.
And here is the part that actually interests me
After more than 20 years in this field: the problem with automated scanners was never the engine. The problem is what they are not capable of seeing.
Business logic flaws. Authorization problems. A flow that lets one user pull another user's data. Nothing about those requests is malformed - there is no pattern to match, because the request is perfectly well-formed and perfectly legitimate. It is only wrong when you understand what the system is supposed to do.
Swapping a rules engine for a language model makes the scanner better at what scanners already do. It does not move the boundary.
I first shared a version of this as a LinkedIn post on 2026-07-18. It is republished here, lightly edited, so it is easier to find and reference. — Erez Metula
Keep reading
More from the blog
AI and LLM Security
Somebody Wired the Darknet Into Your AI. What Could Go Wrong?
One MCP server packing 66 tools, feeding onion-site content straight into your model's context. The darknet is the most hostile input that…
Read itAI and LLM Security
Your AI Coding Tool Walks Straight Past Your Defences
A prompt injection hidden in a README, and the tool writes to your zshenv. Every new shell runs the attacker's code. And…
Read itAI and LLM Security
It Produced Harmful Output 80% of the Time Without Anyone Even Attacking It
An AI system broke Meta's model in 85% of attempts, in three hours, with no code. The part that should worry you…
Read itTell us what the system does and what worries you.
If a penetration test is not what you need yet, we will say so.