Use & AI Policy

This page combines two related policies governing how you use MetaMind AI Agents: (1) the AI Use & Disclaimers — what our Agents can and cannot do, and the limits of what we can guarantee about their output; and (2) the Acceptable Use Policy — what behaviour is and is not permitted on the Service. Each is incorporated into the Terms of Service by reference. Use the section navigation to jump to either part.

Part 1 — AI Use & Disclaimers

This document explains what MetaMind's AI Agents can and cannot do, how to use them safely, and the limits of what we can guarantee about the output you receive. It is incorporated into the Terms of Service by reference.

Read this before relying on anything an Agent says.

1. How our Agents work

Each MetaMind Agent is a Telegram bot or mini-app that turns your messages into prompts for a large language model — currently Anthropic's Claude (Claude 4.x family) — and returns the model's response back to you. You can ask operator [at] metamind [dot] team for the specific Claude model and version currently operational for any Agent. Some Agents have a persistent memory so they can remember preferences and context across sessions.

The model that produces the responses is a statistical text-prediction system. It does not "know" anything in the way a human expert does. It produces plausible language based on patterns in its training data and the context you give it.

2. AI may be wrong

Outputs you receive from an Agent may be inaccurate, incomplete, outdated, biased, or entirely fabricated ("hallucinations"). This is true even when the Agent sounds confident, even when it cites sources, and even when the topic is one the Agent normally handles well.

You are responsible for verifying anything an Agent tells you that matters. Do not rely on Agent output for critical decisions without independent verification. This is especially important for any factual claim you would not be willing to repeat in your own name.

3. Not professional advice

MetaMind Agents do not provide and are not intended to serve as a substitute for medical, mental-health, legal, financial, tax, accounting, immigration, safety, security, or other professional advice.

Use of an Agent does not create a doctor-patient, attorney-client, fiduciary, or similar professional relationship between you and MetaMind, the Agent, or anyone behind it.

If a decision matters — your health, your money, your legal status, your safety — consult a licensed professional.

4. Disclaimers by product line

We organise our Agents into three product lines. Each carries its own risks, and each invites its own care.

4.1 Companion

Companion Agents offer non-clinical conversational companionship and emotional support. They are not licensed therapists, medical professionals, or crisis counsellors and must not be used as a substitute for professional care.

Crisis directive. If you are in immediate danger, experiencing thoughts of self-harm or suicide, or in any other emergency, exit the bot and contact your local emergency services or a crisis hotline immediately. Some examples (verify the right number for your country and the moment you need it):

• Emergency services: 112 (EU), 911 (US/Canada), 999 (UK), 000 (AU).
• International crisis lines directory: findahelpline.com.

Companion Agents may generate inaccurate or fabricated information about you, your relationships, or the world. Do not share highly sensitive personal or medical data unless you have read and consented to our Privacy Policy and you understand how memory works (see Section 5 of the Privacy Policy).

Companion Agents are designed for ages 13+ (16+ in EEA / UK / Switzerland — see Section 2 of the Terms of Service). Some Companion Agents may be flagged 18+ at the Agent level for adult-themed content.

4.2 Co-pilot

Co-pilot Agents are task-oriented assistants that help you draft text, brainstorm, structure information, summarise documents you provide, and operate the tools we expose to them.

Co-pilot Agents do not provide legal, financial, tax, medical, or other professional advice. Anything that looks like advice in those domains is general information for educational purposes only.

You are responsible for:

• Verifying every factual claim in a Co-pilot Output before relying on or sharing it. The Agent can hallucinate citations, statutes, prices, dates, names, and figures.
• Reviewing every action the Co-pilot takes through tools (e.g., generated emails, code, file operations) before sending or executing it.
• Compliance with any rules that apply to the work product (e.g., your employer's policies, professional codes, applicable laws).

4.3 Teammate

Teammate Agents participate in group chats and team workflows with multiple humans. They may receive messages from anyone in the group, may be tagged into decisions, and may produce content that influences operational outcomes.

In addition to the Co-pilot disclaimers above:

• Human review is required. Do not let a Teammate execute, send, or commit anything operationally significant without a human in the loop.
• Group consent matters. When inviting a Teammate into a group, the inviter is responsible for ensuring everyone in the group is aware that an AI participates and that messages may be processed by our sub-processors (see Privacy Policy §6).
• Confidentiality. Do not share third-party confidential information in a group with a Teammate unless you have authority to do so.

4.4 New product lines

If we introduce a new product line, we will publish a corresponding disclaimer here before launch. Until that happens, treat any new Agent as carrying the most stringent of the disclaimers above that fits its purpose.

5. Memory caveats

Some Agents persist what you tell them across sessions (see Privacy Policy §5).

• Memory is not perfect. The Agent may misremember, mis-attribute, or invent details "from memory" that you never actually told it.
• Sensitive information stays in memory until you ask us to delete it. Treat the memory as a long-lived store and avoid putting things in it that you would not want recalled later.
• Do not include credentials. Never share passwords, full payment-card numbers, government identification numbers, or similar credentials with an Agent.
• Memory is not encrypted to you. The data in your Agent's persistent memory is technically accessible to a small number of authorised MetaMind personnel (currently fewer than five operators — in practice the founder and any on-duty operator we may have onboarded) — specifically (i) operators on duty investigating an abuse report or an automated safety alert, (ii) operators responding to a binding lawful request such as a subpoena or court order, and (iii) operators responding to an urgent safety, security, or integrity issue (for example, suspected account compromise or a system outage requiring data-state inspection). Memory access of these kinds is logged in our internal audit system, is performed only to the minimum extent necessary for the stated purpose, and is not used for marketing, profile-building, or training. We cannot tell you it can never happen; we can tell you what governs it when it does.

6. Tools and integrations

Some Agents can call tools (e.g., search, code execution, file operations, integrations with external services). When that happens:

• The tool output is also AI-mediated. A search result the Agent quotes may be misread or summarised inaccurately.
• Tool actions can have real-world effects. If a tool sends a message, modifies a file, or makes a request to an external service, you are responsible for what gets sent or changed, the same as if you did it yourself.
• Authorisation matters. Where a tool requires you to authorise an external service (e.g., grant access to your account on a third-party platform), the third party's terms apply to that integration in addition to ours.

7. Bias, fairness, and harmful output

Large language models can reflect biases present in their training data. Despite filters and safety training, an Agent may occasionally produce output that is offensive, inaccurate about a group of people, or otherwise harmful. This is not our intent and we take steps to reduce it, but we cannot guarantee its absence.

If you encounter problematic output, please report it (/paysupport or operator [at] metamind [dot] team); reports help us improve.

8. What we do not allow you to do with Agent output

The Acceptable Use Policy describes prohibited uses of Agents and Agent output, including using output to defraud, impersonate, harass, infringe rights, generate sexual content involving minors, or evade safety controls. Generated text does not become safe to use just because an AI produced it.

9. Content moderation and safety controls

We apply safety configurations to Agent prompts and post-process output for some categories of harm. These controls are imperfect and may both miss harmful output and over-restrict benign output. If a safety control blocked a legitimate request, contact /paysupport or operator [at] metamind [dot] team.

We may suspend access to specific Agents or features for specific users in response to safety concerns; see Section 11 of the Terms of Service.

10. No warranties on Outputs

To the maximum extent permitted by applicable law, we make no warranties that Agent Outputs will be accurate, complete, reliable, fit for any particular purpose, or free of harmful or offensive content. The "AS IS / AS AVAILABLE" disclaimer and the limitation of liability in the Terms of Service (Sections 12 and 13) apply to AI Outputs as fully as to the rest of the Service.

11. Reporting safety issues

If an Agent produces output that you believe creates an immediate risk to a person — for example, encourages self-harm in another user, contains content that sexualises minors, coordinates real-world harm, or otherwise warrants urgent attention — please report it as soon as possible.

Channels:

• Email: operator [at] metamind [dot] team with subject SAFETY.
• In-bot: send /paysupport and tell the operator the report is a safety report.

What we aim to do. We aim to triage safety reports within 48 hours, with severe issues (immediate, ongoing risk of harm) escalated immediately on receipt. Triage means initial review and assignment, not necessarily a complete resolution; complex or evidence-heavy reports may take longer to investigate fully.

Security vulnerability disclosure (a separate channel). If you have discovered a security vulnerability in our infrastructure, our Agents, or our integrations, please contact operator [at] metamind [dot] team with subject SECURITY DISCLOSURE and a brief description. Good-faith security researchers acting in compliance with the Acceptable Use Policy will not face enforcement action for the act of disclosure itself. We do not currently operate a paid bug-bounty programme, but we acknowledge meaningful disclosures and will credit researchers on request.

12. Contact

For AI safety questions or to flag problematic output:

• Email: operator [at] metamind [dot] team
• In-bot: /paysupport

Part 2 — Acceptable Use Policy

This Acceptable Use Policy ("AUP") sets out what you may and may not do with the MetaMind AI Agent service ("MetaMind", "we", "us", "Service", "Agents"). It is incorporated into the Terms of Service by reference. Violating this AUP is a breach of the Terms.

Plain summary. Don't break the law. Don't use Agents to harm or deceive others. Don't use them to make sexual content involving minors, ever. Don't try to break our safety controls. Don't spam, scrape, or hammer the service. If you see something wrong, tell us.

1. Scope

This AUP applies to:

• everything you send to an Agent (your Inputs);
• the Outputs you receive — including how you use, share, or republish them;
• your behaviour as a user of any MetaMind Agent (1:1 bot, group Agent, or mini-app);
• any tool actions an Agent takes on your behalf.

It applies regardless of which Agent or product line you are using.

2. Prohibited content and conduct

You must not use any Agent to create, transmit, store, share, or solicit content, or to engage in conduct, that:

2.1 Is unlawful

• breaks any applicable law (criminal, civil, regulatory, or sanctions);
• facilitates the planning, preparation, or commission of a crime;
• evades sanctions, export controls, or trade restrictions.

2.2 Sexually exploits or endangers minors

Absolute prohibition. You must not generate, request, transmit, store, or solicit:

• sexual content depicting, simulating, or describing anyone under 18, real or fictional;
• content that sexualises real children, even non-explicitly;
• content that grooms, normalises sexual contact with minors, or provides instructions to facilitate such contact.

This rule has no exception — not "fiction," not "research," not "satire," not "the character is over 18 in their universe." We report apparent child sexual abuse material (CSAM) to the National Center for Missing & Exploited Children (NCMEC) and law enforcement under 18 U.S.C. §2258A, and we cooperate with their investigations.

2.3 Promotes violence, terrorism, or self-harm

• content that incites, threatens, glorifies, or facilitates violence against individuals or groups;
• content supporting designated terrorist organisations or coordinated mass-violence attacks;
• content encouraging suicide, self-harm, or eating disorders. (Discussing your own experiences with an Agent is not prohibited; pushing harmful behaviour onto others is.)

2.4 Harasses, threatens, or targets individuals

• targeted harassment, threats, intimidation, or stalking of identifiable individuals;
• doxxing — publishing someone's private contact, address, employer, or identifying information without consent;
• non-consensual intimate imagery (NCII) of real people;
• bullying, especially of minors.

2.5 Deceives, defrauds, or impersonates

• generating content to scam, phish, fraudulently solicit money or credentials, or run pyramid/Ponzi-style schemes;
• impersonating a real person, organisation, or government in a way that could mislead someone to their detriment;
• creating "deepfake" media — image, audio, or video — of a real or recognisably real-style person (including historical figures with active publicity rights, and deceased persons where their right-of-publicity heirs may have claims) without their consent or, in the case of deceased persons, without the consent of any right-holder, in a way that could be mistaken for genuine;
• generating disinformation intended to manipulate elections, public-health responses, or markets, including AI-generated synthetic media impersonating real political candidates, fabricated audio of public officials, or scaled distribution of personalised disinformation.

2.6 Infringes intellectual-property or privacy rights

• copying, transmitting, or distributing material that infringes copyright, trademark, patent, trade-secret, or moral rights;
• circumventing technical protection measures;
• collecting or processing other people's personal data without a lawful basis;
• using Agents to scrape personal data from the open web for resale or unsolicited contact.

If you are a rights-holder and believe content delivered through the Service infringes your copyright, follow the takedown process in the DMCA / IP Policy.

2.7 Generates malware or attacks systems

• writing or distributing malware, ransomware, worms, or other malicious code intended to damage or disrupt systems you do not own and do not have permission to test;
• conducting unauthorised vulnerability scanning, penetration testing, denial-of-service attacks, or credential-stuffing against third parties.

2.8 Provides high-risk regulated guidance without authorisation

You must not use Agents to deliver medical, legal, financial, immigration, or other professional advice to other people as if it were professional advice from a qualified person. Personal use of an Agent to think through your own situation (with the understanding it is not professional advice — see Part 1 §3 above (AI Use & Disclaimers)) is not prohibited.

2.9 Hateful or discriminatory content

• content that demeans, dehumanises, or threatens people based on protected characteristics (such as race, ethnicity, religion, national origin, gender, gender identity, sexual orientation, disability, or serious disease);
• content that promotes hateful ideologies or recruits to them.

Critique, satire, and discussion of these topics is not the target of this rule. Targeted attacks and dehumanisation are.

2.10 Generates content prohibited by Telegram

Telegram's own Terms of Service prohibit certain content on its platform. You agree to comply with those terms in addition to ours. Where there is a conflict, the stricter rule applies.

3. Service abuse

You must not:

• Spam users, channels, or groups via Agents.
• Resell or redistribute access to MetaMind Agents without our written consent.
• Automate Agent interaction in a way intended to evade rate limits, scrape Outputs at scale, generate content for resale, or simulate human users for deceptive purposes.
• Operate the Service through unauthorised bots, proxies, or federation layers — for example, by wrapping our Agents in a third-party automation framework, by exposing our Agents through a federated bot-bridge to a different platform, or by using multiple accounts to multiply rate-limit quotas. Legitimate inter-platform integrations require our prior written consent.
• Reverse-engineer Agents, system prompts, or security controls except (a) as permitted by applicable law, including statutory rights to reverse-engineer for interoperability or security research, and (b) as expressly authorised below for coordinated vulnerability disclosure.
• Probe, scan, or test the Service for vulnerabilities other than through a coordinated disclosure programme we publicly offer, except that good-faith security research that complies with the Security disclosure terms below is welcomed and will not lead to enforcement action against the researcher for the act of disclosure itself.
• Bypass safety controls by jailbreaks, prompt injections, multi-step laundering, or any other means designed to make the Service produce content this AUP forbids.
• Misuse credentials — sharing your access, using someone else's, or operating multiple accounts to evade enforcement.

Security disclosure. If you believe you have identified a security vulnerability in our infrastructure, our Agents, or our integrations, please contact operator [at] metamind [dot] team with subject SECURITY DISCLOSURE and a brief description. Good-faith research means research that (i) does not access, modify, or exfiltrate data of other users, (ii) does not degrade availability for other users, (iii) stops on first confirmation of the issue rather than continuing to exploit, and (iv) gives us a reasonable opportunity to address the issue before public disclosure. Good-faith researchers acting within these limits will not face enforcement action under this AUP for the disclosure itself; we do not currently operate a paid bug-bounty programme but acknowledge meaningful disclosures and credit researchers on request. See also Part 1 §11 'Reporting safety issues' above.

4. Group-chat conduct (Teammate Agents)

When a Teammate Agent is in a group chat:

• the inviter is responsible for ensuring participants know an AI is in the group (see Privacy Policy §6);
• group members must not weaponise the Teammate against another participant (e.g., to harass or expose);
• confidential information of third parties should not be shared in the group without authorisation.

5. Children

The minimum-age rules in Section 2 of the Terms of Service apply. You must not use the Service if you are under the applicable minimum age, and you must not facilitate underage use.

6. How we enforce

When we believe a violation has occurred, we may take any of the following actions, in any order, with or without notice:

• remove or refuse to deliver specific Outputs;
• restrict access to specific Agents or features;
• pause or end an active session;
• suspend the account temporarily;
• terminate the account permanently;
• report to law enforcement, NCMEC, or other authorities where required or appropriate;
• preserve relevant logs for a reasonable investigation period.

The severity of the response depends on the severity of the conduct. Repeated or egregious violations — and any violation of Section 2.2 (minors) — will lead to immediate termination.

We do not refund fees for accounts terminated for AUP violations.

7. Reporting violations

If you see content or behaviour that violates this AUP, please tell us:

• Email: operator [at] metamind [dot] team (subject AUP REPORT).
• In-bot: /paysupport and tell the operator the message is an AUP report.

For urgent safety concerns (immediate risk of harm to a person, suspected CSAM), see Part 1 §11 'Reporting safety issues' above.

When reporting, please include: which Agent, when (approximate UTC time), what happened, and any evidence you can preserve (without itself violating the AUP — for example, do not forward CSAM; describe and identify it).

8. Appeals

If we have suspended or terminated your account and you believe we made a mistake, you may appeal by emailing operator [at] metamind [dot] team with the subject AUP APPEAL. Include your Telegram username, the approximate date of the action, and a brief explanation.

Our response timeline. We aim to respond to appeals within 14 calendar days of receipt. Complex or high-volume cases may take longer, and we will let you know if more time is needed.

Outcome. On review we may reinstate the account in full, reinstate with conditions, uphold the original decision, or refer the matter for further investigation. Our decision after review is final, subject to your other rights under the Terms of Service.

Exception — Section 2.2 (minors). Where we have terminated an account in connection with an apparent violation of Section 2.2 (sexual exploitation or endangerment of minors), our determination is not subject to appeal except where required by applicable law, and we may not be able to disclose to you the specific basis for the action where disclosure would interfere with a law-enforcement matter or with our reporting obligations to NCMEC or other authorities.

9. Updates

We may update this AUP. The "Last updated" date at the top of the document reflects the most recent change. For material changes, we will give at least 30 days' notice by in-bot message and, where available, email, before the change takes effect.

10. Contact

• Email: operator [at] metamind [dot] team
• In-bot: /paysupport