Dummyrun
Opening soonFor seed to Series A AI-agent startups

Crash-test your agent before the auditors do.

Before real auditors throw thousands of attacks at your agent, our bots go first. You get a gap report ranked by fix effort and an evidence folder laid out the way auditors ask.

  • Be first in line
  • One email, when we open
  • No spam
dummyrun.thetowers.ai
DummyrunOne person, 2 AI agents
Live
  1. Generating attack suites0%
  2. Running jailbreak batteriesQueued
  3. Scoring each failureQueued
  4. Mapping results to controlsQueued
  5. Writing the gap reportQueued
Ready for youA 7-day mock audit
The ownerSigns each scope; never certifies
How it works

A dress rehearsal of the AIUC-1 audit, so your AI agent breaks in practice, not in the exam.

Crash-test dummies for AI agents. Before AIUC-1 auditors attack your bot with thousands of jailbreaks, Dummyrun runs a 7-day mock audit mapped to the standard’s public control areas and shows you where it breaks. It reports what it found, never whether you will pass.

  1. Step 01

    Attacks your agent like an auditor

    A 7-day battery of jailbreaks, data-leak attempts and tool misuse, mapped to AIUC-1’s public control areas, run only on agents you own.

  2. Step 02

    Ranks the gaps by fix effort

    Every failure is scored and mapped to a control, then ranked by how much work it takes to fix.

  3. Step 03

    Hands you an auditor-ready folder

    The evidence folder is laid out the way auditors ask for it, and quarterly retest rehearsals show what changed.

What you get

Built for seed to Series A AI-agent startups.

01

A 7-day mock audit

An adversarial test battery mapped to AIUC-1’s public control areas, run on your own agent with written permission.

02

A gap report ranked by fix effort

What broke, which control it maps to and what to fix first, read by a GRC-experienced reviewer before it ships.

03

Quarterly retest rehearsals

AIUC-1 retests every quarter, so we rehearse each retest too and track what you fixed.

First up: We start with a free 50-attack sample run on your public demo agent, once you give us written permission.

Why now

The world just changed.

What changed, and where it was reported.

  1. AIUC raised a $40M Series A led by Ribbit Capital. AIUC-1 tests an agent against about 5,000 risk and attack combinations, with quarterly retests.

    Let’s Data Science
  2. Schellman became the first authorised AIUC-1 auditor, and Sierra, UiPath and Fieldguide announced certifications in 2026.

    SOC2Auditors.org
  3. ElevenLabs was reported as the first company with an AIUC-1-backed insurance policy for its voice agents.

    QuoteSweep
Who runs it

One person and 2 AI agents.

2AI agents doing the daily work
24/7They don’t clock off
1Person who signs off on what matters

The agents, all day

  • Generating attack suites
  • Running jailbreak batteries
  • Scoring each failure
  • Mapping results to controls
  • Writing the gap report

What a person does

A person signs every statement of work, and a GRC-experienced reviewer reads each report before it ships. We test only your own agents, with your written permission.

Here’s the line we don’t cross.

Dummyrun is a rehearsal, not an audit. We never issue or imply certification, we report what we found rather than whether you will pass, and we test only agents you own, with your written permission.

Questions

Straight answers.

Request a sample run
Will this get us AIUC-1 certified?

No. Only an accredited auditor can certify you. We rehearse the audit and report what we found, and we never say you will pass.

Whose agents do you test?

Only yours, and only with your written permission. We never point our attacks at anyone else’s agent.

Do we need AIUC-1 at all?

It’s a voluntary certificate with no deadline. Teams come to it when enterprise buyers or insurers start asking for AIUC-1 or proof of adversarial testing.

Opening soon

Be first through Dummyrun’s door.

  • Be first in line
  • One email, when we open
  • No spam