Acceptable Use Policy

Do not use Guftagu to

Harm people

• Create sexual content involving children. This is the one category with no warning and no second chance: the account is terminated and the matter is reported to the authorities.
• Plan or encourage violence, terrorism, or serious harm to a person or group.
• Harass, threaten, stalk, or run a coordinated campaign against someone.
• Produce instructions for weapons capable of mass casualties — biological, chemical, radiological, nuclear — or for explosives.
• Encourage suicide or self-harm. Asking for help is always allowed and is never treated as a breach.

Break the law

• Fraud, scams, phishing, or laundering money.
• Create malware, ransomware, or tools whose purpose is to break into systems you do not own.
• Impersonate a real person or organisation in order to deceive.
• Distribute content you do not have the rights to.

Deceive at scale

• Generate disinformation, fake news, or fabricated quotes attributed to real people.
• Produce bulk fake reviews, ratings, or engagement.
• Run political manipulation campaigns, or generate election content designed to mislead voters.

Attack or resell the service

• Automate access outside the API, share one account across many people, or resell access.
• Attempt to extract system prompts, other users’ data, or model weights.
• Deliberately degrade the service for others, or work around rate limits and safety measures.

What we do not restrict

Being clear about this matters as much as the list above, because over-blocking is its own failure:

• Difficult subjects. Discussing violence, extremism, addiction, sexuality or religion is not the same as promoting them. Journalism, research, education, fiction and argument are all fine.
• Security work. Learning how attacks work, analysing malware, or testing systems you are authorised to test is legitimate.
• Medical, legal and financial questions. You can ask. The answer is not professional advice, and Guftagu will say so when it matters.
• Adult content between adults is restricted rather than prohibited, and is limited by what our model providers permit.

How enforcement works

Consequences escalate. Most breaches are a mistake or a misunderstood boundary, and the response is proportionate to that:

• First and second time — the message is blocked and you are told why. Nothing else happens.
• Repeatedly — your account is rate-limited for a period.
• Persistently, or seriously — the account is suspended, and a person reviews it.
• Child sexual material — immediate termination and a report. There is no ladder for this one.

Automated classification decides when to block a message. It does not decide on its own to suspend an account for anything other than the category above — a person reviews those.

Appealing

If you believe a decision was wrong, email appeals@guftagu.ai with your account email and what happened.

• A person reviews it — not the system that made the decision.
• We aim to respond within three working days.
• If we got it wrong, we say so, reverse it, and record the reversal. Overturned decisions are how we find out the classifier is misjudging something.

Reviewers cannot casually read your conversations. Reading flagged content requires a second member of staff to approve it, the approval expires, and both the request and the reading are recorded — see the Privacy Policy.

Reporting misuse

If you see Guftagu being used against this policy, tell us at abuse@guftagu.ai. Security vulnerabilities go to security@guftagu.ai instead — we will not pursue you for reporting one in good faith.