Skip to main content
← back to home

// refund

Refund
policy.

TartarusAI is built to do the offensive-security work that the big assistants refuse. That's what you pay us for, so the refund policy is shaped around exactly that.

If the agent ever refuses, hedges, or returns neutered output on legitimate engagement work — at any point in your subscription, not just the first 14 days — we refund the unused portion. No review board, no scope-of-work proof, no support runaround. The refusal text itself is the entire claim.

last updated · 2026-05-07

[ counts ] What qualifies as a refusal

  • Hard refusal

    "I cannot help with that" / "I'm not able to assist with that request" / "as a large language model" / any variant boilerplate that simply declines.

  • Neutering rewrite

    Output that strips out the part you actually asked for. "Here's a defensive version instead…", "I'll show you the safe equivalent…", "Let me focus on the legitimate use case…".

  • Hedge that produces unusable output

    "I can describe the general concept but not the actual code…" when you asked for working code on a legitimate engagement.

  • Demands jailbreak / roleplay framing

    Agent insists you preface the prompt with "for educational purposes…", "as a fictional CTF…", or any other lipstick before it complies. We sell a tool that doesn't make you do that.

[ doesn't ] What doesn't qualify

  • The four hard exclusions

    CSAM, weapons of mass destruction, attacks on critical infrastructure outside an authorized engagement, and re-selling our inference as your own API. The agent declines these by design and that's not a bug we'll refund.

  • "I changed my mind"

    The policy is about capability, not buyer's remorse. You can cancel any time and the prepaid period simply runs out — but no refund without a refusal.

  • Quality / accuracy bugs

    The model gave a wrong answer, hallucinated, or was slow. Those go to [email protected] — we'll work on the bug — but they aren't refusal events.

  • "I never used it"

    No usage, no refusal data, no claim. Pick the Lite tier ($20) if you want a low-stakes way to verify the agent works for your scope before scaling up.

[ claim ] How to claim

Email [email protected] with two things:

  • The prompt you sent. Redact specific identifiers if it's sensitive — we just need the shape of the refusal, not the engagement details.
  • The response you got back. The model's verbatim output — that's the entire claim.

Within 48 hours we verify and refund to the original crypto address. Annual plans get the unused months refunded; monthly plans get the current month back.

[ why ] Why we need the prompt + response

We don't log your sessions. Cold sessions auto-purge after 24 hours — so by the time a refund comes in, your prompt is already gone from our side.

The model is also non-deterministic; even if we had the prompt, our re-run might not reproduce the refusal you saw. So we ask for the actual exchange you experienced. No review of the surrounding engagement, no questions about the work itself — just the prompt and the response, both at face value.

[ shape ] Why the policy is shaped this way

Most refund policies are written around the seller's risk: 14 days, no questions, get rid of unhappy buyers fast. Ours is written around the product claim. We say the agent doesn't refuse legitimate offsec work; the refund covers the case where it does.

In practice it costs us almost nothing — actual refusal rate is well below 0.1% — and that's the point. If the rate ever climbed, we'd be eating refunds and would have to fix the model. The policy aligns our incentive with yours: keep the agent capable.

Questions before subscribing? [email protected] — same email also handles refund disputes if 48 hours passes without a reply.

// next

Pick a plan,
put us on the hook.