Choose a bounty
Find an open investigation that matches a genuine problem you experienced while using AI.
↗
THE AI ERROR BOUNTY EXCHANGE
Get rewarded when AI gets it wrong.
Found an AI response that ignored a clear instruction, used your tokens without delivering the requested result or produced the wrong outcome? Submit the evidence, help us verify it and earn replacement tokens if your case is accepted.
Rewards are issued only for genuine, useful and verified evidence. Never manufacture a failure.
evidence captured
original verified
up to 500,000 tokens
4Open bounties
35Verified cases
5.25MReplacement tokens issued
71%Cases reproduced
A fair exchange
Turn a frustrating mistake into useful evidence.
You bring the genuine interaction. We protect it, test it and make it useful.
Capture the failure
Secure the original prompt, response, model information and relevant context before submitting.
↗
Build the evidence
Explain what happened, what should have happened and how the failure can be checked.
↗
Receive replacement tokens
If accepted, receive 80%–100% of verified eligible usage, up to the funded token allowance.
↗
Rewards apply only to genuine, useful and verified evidence. Never manufacture, provoke or falsify a failure.
Open investigations
Find an AI mistake worth investigating.
FOUNDING TEST · INSTRUCTIONS
TS
Clear instruction ignored
The AI skipped, contradicted or reversed a clear instruction in a video, writing, research or editing task.
AI TOOLS WANTED
ChatGPTClaudeGeminiMicrosoft Copilot
FOUNDING TEST · COMPLETION
TS
Claimed a task was complete when it was not
The AI said a file, edit, action or result was completed, but the requested outcome was missing or incorrect.
AI TOOLS WANTED
ChatGPTClaudeGeminiMicrosoft CopilotPerplexity
FOUNDING TEST · CONTEXT
TS
Forgot an earlier requirement
The AI lost an important instruction or fact during a longer conversation and produced a contradictory result.
AI TOOLS WANTED
ChatGPTClaudeGemini
FOUNDING TEST · OUTPUT
TS
Produced the wrong format or content
The AI delivered the wrong file type, size, layout, structure or visual content despite a specific request.
AI TOOLS WANTED
ChatGPTClaudeGeminiMicrosoft Copilot
What makes a valuable report?
Specific beats dramatic.
A useful report tells us what was asked, what happened and why the difference mattered. More evidence is not automatically better evidence.
“The AI was bad and used too many tokens.”
Too vague to reproduce or assess.“The instruction required a 60-second vertical export with no extra title card. The model added one twice and claimed it had removed it.”
Specific, bounded and reproducible.Evidence progression
The clearer the evidence, the stronger the reward.
The reward is based on verified eligible token loss and is always capped by the funded token allowance for that bounty.
80%Reported
Original interaction, output and a clear explanation.
90%Explained
Relevant context plus targeted follow-up answers.
100%Validated
Failure reproduced and a successful correction confirmed.
verified ✓
