$9,000 · 3 weeks

AI Security Review

A fixed-scope review of where an AI system in your product can be attacked, and what happens if it is. Ranked findings, hardened examples, and a 90-minute readout that shows your team what to fix first and why.

Price
$9,000
Timing
3 weeks
Best for
An AI feature already shipping or about to
Deliverable
Ranked findings, hardened examples, 90-minute readout
A left-to-right threat path from untrusted input through model context and an enforceable approval gate to a protected system.
Follow the path from untrusted input to the action your system can actually take.

What this finds

Where untrusted input reaches a model that can call tools, and what an attacker gets by controlling them: agent permission blast radius, data flow to model providers, tenant isolation, output handling, and whether your human-in-the-loop gate is actually enforceable, not just documented.

What you get

Ranked findings, hardened examples for the issues that matter most, and a 90-minute readout with your team. A written record, not a slide deck that gets forgotten.

Who this is for

A team that already has an AI feature, such as a chat surface, an agent that calls tools, or a pipeline that hands a model access to internal data, shipping now or about to. You want a technical answer, not a checklist.

Who this isn’t for

If you know AI security questions are coming but don’t yet have anything built to review, start with the Roadmap Sprint instead. It’s a week, not three, and it’s built for exactly that stage.

Questions

What if the review finds nothing serious?

That happens, and it’s a real result. You get a written record that the system was checked and what was checked, which is exactly what a security questionnaire or a board asks for.

Do you need source code access, or is this a black-box test?

Source access, architecture diagrams, and IAM/policy configuration. I read the system, not just poke at its edges. Findings are more concrete and defensible that way, and it’s scoped on the intro call before we start.

What’s not included?

No infrastructure changes or merged code. You receive findings and hardened examples you or your team implement. If you want implementation help after the review, that’s a separate conversation, not part of the fixed scope.

Can this run under NDA?

Yes. I’m happy to sign yours before we talk specifics, as with any engagement.