Disclaimer & terms.

Funding and payout pool

Cash-back payments come from a limited HugClaims demo fund, sponsor pool, or research budget. Payouts are not guaranteed until a claim is accepted, verified, and marked payable.

• Payout amounts are capped. The displayed amount is an estimate based on stakes, confidence, available funds, and review status.
• Funds are distributed first to valid, non-duplicate claims. If the pool is exhausted, otherwise valid claims may receive a reduced payout, deferred payout, or no payout.
• High-stakes does not automatically mean payable. The correction still has to identify a real AI error and provide a substantially right fix.
• Corpus saturation can pause payouts. If a failure corpus is sufficient, we will not accept new claims for cashback for that failure type.

One claim per AI error

Each distinct question, AI answer, and alleged error can be submitted once. Rewording the same prompt, resubmitting the same transcript, or splitting one mistake into many small claims may be treated as duplicate activity.

• One question, one wrong answer, one correction. If several people submit the same underlying error, the earliest complete valid claim normally has priority.
• Follow-up claims need a new error. A later claim must show a materially different AI answer or a different substantive mistake.
• Forum examples are not bounty templates. Copying public examples without adding a new verified error is not payable.
• Daily submission cap applies. Each account is limited to 15 claim submissions per calendar day.

What counts as a valid claim

• The AI answer made a factual, mathematical, logical, citation, reasoning, or materially misleading error.
• The submitted correction is substantially right, specific, and supported by the conversation or credible evidence.
• The error matters enough to affect the answer, not merely style, tone, formatting, harmless ambiguity, or preference.
• The transcript, screenshots, links, or evidence are authentic and complete enough for review.

Abuse we may reject

• Manufactured errors: prompts designed only to trick, coerce, jailbreak, or force a known failure without a real user need.
• Duplicate farming: submitting the same issue through multiple accounts, brands, paraphrases, or near-identical transcripts.
• Collusion: coordinating with reviewers, forum users, or other claimants to inflate votes, agreement, or payout priority.
• Bad corrections: replacing an AI error with another incorrect answer, speculative claim, or unsupported assertion.
• Unsafe or illegal content: claims involving instructions for harm, evasion, fraud, privacy invasion, or prohibited activity may be refused or redacted.
• Evidence tampering: edited screenshots, omitted context, fake transcripts, or altered model outputs disqualify the claim.

Review and appeals

Automated review is only a first pass. HugClaims may use human review, community signals, model-based verification, or source checking before payment. We can reverse, reduce, or pause a payout if later evidence shows the claim was duplicate, invalid, abusive, or unsafe.

A verifier saying "valid" is not a final legal or financial commitment. It is a review signal used by HugClaims.

Privacy and publication

Submitted conversations may contain sensitive information. Do not submit private, confidential, medical, legal, financial, employer, client, or third-party data unless you have the right to share it. Forum posts may be visible to other users and may be summarized for research or product feedback.

How submitted data may be used

By submitting a claim, importing a chat, attaching evidence, posting to the forum, or using the verifier, you allow HugClaims to process that content for operating, reviewing, improving, and researching AI-error detection systems, subject to applicable terms and privacy rules.

• Review and verification: we may use submitted prompts, AI responses, corrections, evidence, votes, comments, and verifier outcomes to decide whether a claim is valid, duplicate, abusive, payable, or needs human review.
• Model improvement: we may use submitted AI-error examples to train, evaluate, fine-tune, benchmark, or otherwise improve better LLMs, verifiers, reward models, classifiers, retrieval systems, and safety tools.
• Research datasets: we may transform submissions into aggregated, redacted, or structured datasets for studying where AI systems fail and how corrections should be judged.
• Public forum content: if you choose to share a claim publicly, the post, comments, votes, and visible claim context may be used as public feedback signals for community review and model-quality research.
• Terms apply: model-training and research use may be limited by law, platform terms, data-provider terms, sponsor requirements, privacy commitments, opt-out settings, and any written agreement that applies to your use of HugClaims.

Do not submit secrets, private documents, personal identifiers, client data, or regulated information unless you have permission and are comfortable with the data being reviewed and potentially used for AI-quality research.

Not insurance, advice, or a warranty

HugClaims is not an insurance company, regulated financial product, legal service, medical service, or guarantee that an AI system is safe or correct. Cash back is a reward mechanic, not coverage for real-world losses.

For medical, legal, financial, safety, or other high-stakes questions, consult a qualified professional and verify independently.