Acceptable use policy

Last updated: 4 October 2026. Draft of 4 October 2026.

This policy is part of our terms of service. It applies to everyone who uses Pearl Compute: API customers, chat users, and GPU operators in our miner program. If you build our API into your own product, you are responsible for making sure your users follow it too.

1. What you may not do with the models

Do not use Pearl Compute, directly or through your own product, to:

• break the law, or help anyone else break it;
• create, request or share sexual content involving minors, in any form. There are no exceptions;
• create sexual or intimate content of a real person without their consent;
• create malware, or plan or carry out attacks on computer systems you are not allowed to test;
• harass, threaten, bully or intimidate people, or encourage violence against them;
• promote terrorism or violent extremism, or encourage self-harm or suicide;
• defraud or deceive people: scams, phishing, spam, fake reviews, or impersonating a real person or organisation to mislead;
• create or spread disinformation meant to deceive the public, including about elections;
• help develop or use weapons, including chemical, biological, radiological or nuclear weapons;
• collect, process or reveal personal data unlawfully, for example tracking or profiling people without a legal basis;
• make decisions that have legal or similarly significant effects on people, for example about credit, jobs, housing, education or access to essential services, based only on model output without meaningful human review;
• carry out practices that the EU AI Act prohibits, such as social scoring or manipulating people in ways that cause them harm;
• violate other people's rights, including intellectual property and privacy rights;
• use a model in a way its licence or use policy forbids (section 4).

2. What you may not do to the service

Do not:

• try to get other users' data, including their requests, outputs, account details or usage, by any means, including side channels such as timing or traffic analysis;
• attack, overload or disrupt our website, API, gateway, mining pool or GPU nodes;
• get around rate limits, authentication, billing or other limits, for example by spreading traffic across accounts to avoid limits;
• probe or test our systems for vulnerabilities without our written permission. If you find a security problem, tell us at security@pearlsafe.xyz and give us time to fix it before you share it;
• use someone else's account or API key, or publish your own key where others can use it;
• abuse the free allowance, for example by creating several accounts, automating the chat, or using it as a free API;
• resell, sublicense or give others access to the free allowance or the web chat.

You may build the paid API into your own products and offer them to your customers, including by reselling access to the paid API, as long as you and they follow this policy.

3. Rules for miners

If you run a GPU node or mine with our pool, you must also not:

• withhold shares or blocks, for example submitting ordinary shares but holding back a share that would win a block, or delaying shares;
• manipulate shares or rewards: resubmit or duplicate shares, submit shares that did not come from the work we assigned, misreport what your node was doing when a share was found (customer traffic, filler or idle), or split one machine into many workers to game the share difficulty;
• misreport work: serve a different model from the one you claim, change token counts or usage receipts, or handle our test requests differently from customer requests;
• misreport your node's location: declare a false country, or hide the node's real location (for example behind a VPN or proxy) to pass our country check;
• modify our node software or miner image, or change its settings, including its logging settings;
• log, store, inspect, copy or share customer content, or run anything on the node that can read it;
• attack other miners, the pool or the Pearl network.

4. Model licence restrictions

Each model's licence and use policy apply to your use of that model and its output. The current list is on our model licences page.

• Llama 3.1 8B Instruct: Llama 3.1 Acceptable Use Policy and Llama 3.1 Community License. Built with Llama.
• Gemma 4 31B IT: Gemma Prohibited Use Policy. Google publishes Gemma 4 under the Apache License 2.0. The version we serve is Pearl Research Labs' quantized copy of Gemma 4. Whatever licence applies to that copy, you must follow the Gemma Prohibited Use Policy when you use Gemma models through our service.
• Qwen3 30B A3B Instruct (not offered yet): the Qwen team at Alibaba Cloud publishes it under the Apache License 2.0. You must follow its licence terms when we offer it.

When we add a model, its licence and use policy apply from the day we offer it.

5. Enforcement

What we can see. We do not log or store the content of requests, so we cannot review it after the fact. We act on:

• reports from users, rights holders and authorities;
• metadata, such as request volumes, rates, error patterns and payment problems;
• signals from our pool and nodes, such as invalid shares and failed test requests.

What we may do, depending on how serious the breach is:

• warn you;
• lower your rate limits;
• revoke API keys;
• suspend or close your account;
• stop routing requests to a node, ban a mining worker or operator, and hold back payouts linked to the breach (section 12.6 of the terms);
• report to the authorities where the law requires it, or where people are at risk.

We usually warn you first. For serious breaches, such as sexual content involving minors, attacks on the service or share manipulation, we may act at once without notice.

If you think we got it wrong, email legal@pearlsafe.xyz. A person will review the decision and reply within 14 days.

6. Reporting abuse

Email legal@pearlsafe.xyz. If you have them, include request IDs and times. We can look up the metadata of a request, but not its content. Report security problems to security@pearlsafe.xyz.