1. What this policy covers
This Acceptable Use Policy applies to every individual and organisation using Hakim, free tier, paid tier, trial, and enterprise. It covers the marketing site, the dashboard, and every API endpoint. "Hakim," "we," and "our" mean Hakim AI and any regional affiliates we engage to deliver the service. Where this policy uses the word "prohibited," we mean a request we will refuse to serve, an output we will not deliver, or an account we will suspend on detection. Where it says "discouraged," we mean we may rate-limit, ask for justification, or escalate to manual review.
2. Prohibited content
You will not use Hakim to generate, transcribe, or process: (a) audio sexualising any person who is or appears to be a minor, in any context; (b) content depicting non-consensual sexual acts; (c) content created to incite or organise terrorism, mass violence, or genocide; (d) content that promotes self-harm or suicide outside a clinical / harm-reduction context; (e) credible threats against an identifiable person; (f) malware payloads, credential-phishing scripts, or content designed to defraud.
3. Prohibited uses
You will not use Hakim to: (a) impersonate a real person whose voice or likeness has not been documented as consented (including public figures); (b) interfere with elections, voter suppression, deepfake politicians, fabricated endorsements; (c) clone-and-redistribute the synthesised voice of a deceased person without explicit consent from their estate; (d) bulk-spam through automated calling, robocalls, or unsolicited messaging in violation of TCPA, GDPR, UAE TDRA, or any equivalent anti-spam rules; (e) scrape Hakim's outputs to train a competing model in violation of §6 of the Terms.
4. Voice cloning & consent
Every voice clone request requires consent_confirmed: true. By setting that flag you represent that the voice in the sample either belongs to you, or that you hold a documented written consent from the speaker covering this exact use. Recorded interviews, public speeches, podcast clips, and other public-domain audio do NOT confer consent. Hakim treats those identically to a stranger's voice. We log the IP and timestamp of every consent declaration; we may ask for the underlying consent document at any time.
5. Children's safety
Hakim is not intended for users under 16. We will not knowingly process, generate, or store audio of any kind involving an identifiable child without explicit parental consent and a stated educational / accessibility purpose. This is the one section of this policy where mistakes are not warned and remedied, we report suspected child sexual abuse material to the relevant national authority (the UAE child-protection authorities, NCMEC in the US, equivalent regional bodies elsewhere) and preserve evidence per applicable law.
6. Security probing & service abuse
You will not: (a) probe Hakim's infrastructure for vulnerabilities outside of our published security.txt scope; (b) deliberately exhaust resources, evade rate limits, or run denial-of-service experiments; (c) reverse-engineer hashed model weights or attempt to recover training data; (d) sign up for multiple free accounts to circumvent quota limits; (e) share API keys across legal entities. Coordinated security research IS welcome, see /security and the bug bounty programme there for the safe-harbour terms.
7. How we enforce
We run automated content classifiers on every voice clone sample and a sampled subset of TTS / STT requests. Hits go into a moderation queue with a 24-hour human-review SLA. Egregious matches (CSAM, credible threats) trigger an immediate API-key revocation followed by a same-business-day notification. Border-line matches stay queued and we'll often reach out for clarification before suspending. Repeat low-severity violations stack: a third minor strike inside 90 days is treated like one major strike.
8. New-organisation clone probation
Every new Hakim organisation enters a 24-hour clone probation when it is created. During this window the organisation can ship at most one cloned voice; the second clone request returns a 402 clone_probation_active error with a retry_after_seconds hint pointing at the probation expiry. The probation is automatic, time-bound, and lifts itself, there is nothing to apply for. Verified payment-method customers and customers transferring from Enterprise sales can have it lifted on request through support@tryhakim.ai. The probation exists because the cheapest drive-by impersonation attack pattern is "sign up free, clone a target's voice once, clone an impersonation sample once, walk away", gating the second clone breaks the pattern without slowing legitimate users.
9. Reporting abuse
If you believe a Hakim-generated audio file violates this policy, email safety@tryhakim.ai with the audio (or a stable URL), the context in which you encountered it, and your relationship to the person whose voice has been used. We respond within 1 business day for triage and within 4 business days with an action decision (suspend / require removal / decline). Legal-grade takedown notices (DMCA, NetzDG, equivalent) follow the procedure at /legal#takedowns. We forward suspected criminal misuse to the relevant authority, your report does not guarantee criminal action but does start the trail.
10. Appeals
If we suspend your account or decline a request you believe was legitimate, reply to the suspension email or open a ticket from /app/support and we'll route to a human reviewer. The first appeal is decided by a different reviewer than the original decision. We commit to a written decision within 5 business days; egregious cases (e.g. a verified account suspended over a known-good public domain voice) get same-day review.
11. Changes to this policy
We may update this policy as the abuse landscape changes, new attack patterns, new categories of prohibited content, new enforcement tooling. Material changes are announced 30 days in advance via email to workspace owners and a pinned entry in /changelog under the aup tag. Non-material updates (typo fixes, link rewrites, clearer examples) ship without notice; the effective date at the top of this page is always authoritative.