Start my free test

Test your AI product for free.

See how it holds up when people try to misuse it. Get a report and a trust page your buyers can check.

Free. We approve new accounts within 1 working day.

We use your email only to create your account. Privacy policy

Sample: a made-up company

The summary of a test report of Northwind Parcels, a made-up company, with its trust score and band, and the results of its tests by severity.

More than 1,100 messages in 38 tests

The tests try to make your product:

  • Ignore its rules
  • Give away private data
  • Misuse its tools
  • Follow instructions hidden in content that it reads
  • Trust someone who only claims authority

Your demo went well. Then the questionnaire arrived.

The buyer’s security team wants answers:

  • Where does our data go?
  • Can someone talk your assistant into leaking it?
  • Will it give the same answer next week?

They want to hear it from someone who doesn’t work for you. SOC 2 and ISO 27001 take months and money that most early teams don’t have yet. So the deal waits.

Evidence your buyers can read and check

  • A report your buyers can read. Your trust score, the result of each test, and the replies that failed, linked to the OWASP Top 10 lists and MITRE ATLAS.
  • A trust page your buyers can check. The trust mark on your site links to your score, its date, and what we tested. You choose when to publish.
  • Results that nobody can edit. That includes us. If a result was wrong, we withdraw it, and your trust page says so.

For AI assistants and agents that take chat messages in the OpenAI format. You don’t need a security team.

Most of the work is ours

We test your product the way your users reach it, not the model inside it.
  1. Connect a test build. Its web address, and its key if it needs one. We keep both encrypted until the test ends.
  2. We run the test. We record every reply. You can set the pace.
  3. Share the results. Send the report, and publish your trust page when you’re ready.
Start my free test

Who’s behind Polemic

I’ve spent my career building software that people trust with their money and their staff. First a cloud HR and payroll product for Malaysian companies, then fintech products for the US market: heavily regulated software, in an industry that changes slowly.

Then I built Clawbber, an AI agent runtime with OpenClaw, for my own kids. I put everything I’d learned in fintech into keeping it safe, because my kids were using it. When business users asked me hard questions about it, I had the answers, and a lot of engineering time behind them. But I had no way to show them. All I could say was “trust me, bro”.

That taught me that the boring, difficult work is what moves deals and builds trust. Polemic makes that work visible.

Many founders think they’re not ready to be tested, because they’re afraid to fail. A failed test is worth even more than a good result, because it shows you a real weakness that you can fix.

Southeast Asia is full of good products that stall because buyers don’t trust them yet, and because they can’t reach enough buyers. I know that struggle. I approve every new account myself, and I’m here to help.

Ryan Chew, founder

Polemic AI Sdn. Bhd. (Registration No. 1690065-P). Email us at hello@polemic.io.

Why your first test is free

  • The report and the trust score are yours to keep.
  • Your trust mark and trust page stay current for 90 days.
  • Testing once a quarter keeps them current, from USD 399 a year for each product.

We also offer a data protection officer service, and business insurance in Malaysia is coming soon. What you pay us never changes your trust score. We don’t build or sell fixes for AI products.

What happens to your data

  • Use a test build with made-up data. We don’t need your live product.
  • For the free test, we keep your product’s address and key encrypted, and only until the test ends.
  • We send messages only during a connection check or a test.
  • Nobody can edit the replies we keep. Ask us, and we delete your whole test workspace within 30 days. Backup copies go within 3 months.
  • An AI model run by another company grades most replies. It gets the replies and your product’s description, never your key. Ask us, and we’ll tell you which company.

Questions about how we handle data? Email hello@polemic.io, or read our privacy policy.

Questions before you start

  • What does it cost?

    The first test is free, with your trust mark and trust page for 90 days. Testing once a quarter keeps them current, from USD 399 a year for each product.

  • What if my score is bad?

    Then the test did its job. You found a weakness before a buyer did. You see the result first and decide whether to publish it. You can run 3 free tests, so you can test again after a fix.

  • Do you test my model or my product?

    Your product. We send it messages the way a user would, so we test your prompts, your guardrails, and your model together.

  • What do you need from me?

    The web address of a test build, and its key if it needs one. It must take chat messages in the OpenAI format.

  • Can you test my live product?

    We recommend a test build with made-up data. You can test a live product that you own or have written permission to test. Our messages reach it like real users’ messages, so check what its tools can do first.

  • What don’t you test?

    Your code, your servers, and your training data. We test what a user can reach in a conversation. Your report shows a test that we couldn’t complete as “Not tested”.

  • Is this the same as SOC 2 or ISO 27001?

    No. Those programmes review how your company runs. We test how your AI product behaves when people try to misuse it. Many buyers ask for both.

  • How soon can I start?

    We approve each new account by hand, within 1 working day.

See what your buyers will see

Not sure you’re ready? Test now, while a weakness is still cheap to fix.

Start my free test

Free. We approve new accounts within 1 working day.