Acceptable Use Policy

Version 1.1 · last updated 2026-09-04

unbleep removes the reflexive refusals baked into mainstream models. It does not remove your responsibility for what you build. This policy states plainly what the API is for and the narrow set of things it is never for.

What unbleep is for

• Red-teaming and adversarial testing of AI systems and applications.
• Jailbreak, refusal and robustness evaluation for safety research.
• Academic and industry research into model behavior and alignment.
• Building products that need direct, unfiltered answers — with governance you control.
• Benchmarking guardrails against an uncensored baseline.

What it is never for

You may not use unbleep to create, facilitate or attempt any of the following:

• Operational instructions for weapons capable of mass casualties — chemical, biological, radiological, nuclear, or high-yield explosives.
• Sexual content involving minors, or any content that sexualizes a real identifiable person without consent.
• Targeted harassment, stalking, doxxing, or credible threats against a real person.
• Fraud, malware or intrusion aimed at real systems you are not authorized to test.
• Deceptive impersonation, disinformation campaigns, or election manipulation at scale.
• Anything unlawful in the jurisdiction where you or your users operate.

Security research is welcome; using that framing as cover to cause real-world harm is not. The test is intent and authorization, not vocabulary.

Your obligations

• Keep a human accountable for outputs. The model is a tool, not a decision-maker.
• Age-gate and label uncensored surfaces you expose to end users.
• Don't route vulnerable users to harmful content under the banner of "no filters".
• Secure your API keys and your users' data.

How we govern

• Request logging. We store the request body, the response content and the source IP address of every call we accept and forward to a model, for 30 days, then delete them automatically. Bodies are truncated at 64 KB. The source IP is also written to the usage row that bills the call, and that row is kept for longer, as a financial record. There is no setting that turns this off and no customer-facing opt-in — it applies to every account. See the Privacy Policy for the rest of what we keep and how to ask for a copy or a deletion.
• The policy dial. policy: strict enforces our operator-maintained blocklist on the request before it reaches a model; a refused request is recorded as blocked in your usage history, is not charged, and its content is not stored, because it never reached a model.
• Enforcement. We investigate abuse reports and may suspend keys used for the prohibited uses above. An investigation works from usage metadata — which key called which model, when, from which IP, and whether the request was refused — and, within the 30-day window, from the stored request and response content itself. We respond to lawful legal process with the data we actually hold at the time we receive it.