ZexirAI · AI Accuracy & Reliance

Useful output still needs judgment.

This policy explains why generated answers can be wrong, how search and citations should be checked, what safeguards are required for high-stakes work, and when ZexirAI must not be treated as a substitute for a qualified person.

Core rule: Never rely on an AI answer merely because it is fluent, detailed, confident, or accompanied by a citation. Confirm important facts with current primary sources and an appropriately qualified human.

1. Purpose and application

This policy applies to text, summaries, analysis, translations, citations, code, images, audio, search-assisted answers, recommendations, project artifacts, and other output generated or transformed through ZexirAI (“Output”). It is part of the Terms of Service. It applies whether a response comes from a model chosen by you, a model selected through automatic routing, a fallback model used for reliability, or a combination of models, tools, and external information sources.

ZexirAI is designed to help people think, write, research, and build. It is not designed to make final decisions for you. Output is probabilistic: a model predicts a useful continuation based on instructions, context, and learned patterns. It does not necessarily retrieve a verified fact, understand the real world, know your full circumstances, or distinguish a plausible statement from a true one. Product features that sound conversational do not change that technical limitation.

You must apply a level of review proportionate to the foreseeable harm. A casual brainstorming prompt may need little checking. A medication instruction, legal filing, production deployment, financial transaction, employment decision, or public safety statement requires substantially more. If reliable verification is unavailable, do not proceed based on Output alone.

2. Known limitations of generated output

Output may be false, incomplete, outdated, internally inconsistent, overly broad, too narrow, or irrelevant. A model may invent people, events, quotes, cases, statutes, studies, file contents, URLs, product features, commands, or citations. It may perform arithmetic incorrectly, confuse dates or units, omit exceptions, misunderstand negation, assume facts not in evidence, lose track of instructions, or give a polished explanation for a premise that is itself wrong.

Context windows and formatting systems have limits. Long conversations, large files, complex projects, or numerous attachments may be truncated, summarized, sampled, or only partly considered. Scanned text can be misread. Images can be interpreted incorrectly. Audio can be mistranscribed. Search results can be incomplete, ranked for reasons unrelated to truth, or changed after retrieval. A provider interruption can trigger a fallback model with different capabilities.

Models may agree with a user’s framing, mirror an emotional premise, or express uncertainty poorly. Confidence, length, technical vocabulary, citation style, and a professional tone are not measures of correctness. Repeating a question may produce a different answer; consistency across repeated generations is not proof. A refusal or warning is not proof that a request is illegal, and the absence of a warning is not proof that it is safe or lawful.

3. Model selection, routing, and personalization

ZexirAI may offer direct model choices and automatic routing. Routing can consider the selected experience, task type, file or image compatibility, plan access, system availability, reliability, and safety. A requested model may be unavailable or may not support a required input. In those cases the Service may use a compatible fallback or return an error. Provider models change over time, sometimes without identical behavior across versions.

Personalities, intelligence settings, shared memory, conversation history, and age-appropriate instructions influence style and context; they do not make Output verified. Shared memory may be stale or incomplete. A model can misapply a saved preference or infer something you did not intend. Review personalized details and correct the conversation when assumptions are wrong. Do not treat personalization as identity verification or as a dependable record of prior facts.

Different models have different strengths, failure patterns, knowledge cutoffs, tool access, and safety behavior. Comparing models can help surface uncertainty but does not replace source verification. When exact model provenance matters for audit, regulation, or reproducibility, record the displayed model, time, input, relevant settings, sources, and final human review outside the chat.

5. A practical verification standard

Before relying on important Output, identify what would happen if it were wrong. Separate verifiable facts from suggestions, opinions, assumptions, and generated examples. Ask for uncertainties and counterarguments, but verify those responses too. Check names, dates, numbers, calculations, quotations, references, version requirements, and units against primary sources. Recalculate material figures independently and test edge cases.

For work distributed to others, keep a review trail appropriate to the risk. It may include the prompt, source material, generated draft, revisions, test results, reviewer identity, and approval time. Remove fabricated citations and unsupported claims. Confirm that confidential or personal information was not introduced accidentally. Label simulations, placeholders, synthetic media, and invented examples so recipients are not misled.

Use domain-specific tools when they are better suited than a language model. A calculator or tested program should verify arithmetic; a compiler, linter, test suite, sandbox, and security scanner should evaluate code; original records should confirm a summary; and a licensed professional should review regulated advice. When two credible sources conflict, do not conceal the conflict. Describe it and defer the decision to an authorized person.

6. High-stakes and professional contexts

ZexirAI is not a substitute for a physician, therapist, pharmacist, lawyer, accountant, financial adviser, engineer, emergency dispatcher, safety officer, or other qualified professional. Output does not create a professional-client relationship, privilege, fiduciary duty, diagnosis, prescription, legal opinion, investment recommendation, or certification. Professional rules and confidentiality duties may restrict what you are allowed to upload to an AI service.

Do not use Output alone to diagnose or treat a condition; change medication; determine whether symptoms are an emergency; file a legal document; waive a right; sign a contract; trade or allocate assets; calculate tax owed; make structural, electrical, chemical, aviation, automotive, cybersecurity, or industrial-safety decisions; or control equipment capable of harming people or property. Contact local emergency services when there is immediate danger. Verify phone numbers and procedures from an official local source.

Output must not be the sole basis for decisions that determine another person’s access to employment, housing, education, credit, insurance, health care, legal services, essential utilities, immigration benefits, or other high-impact opportunities. Such uses can be inaccurate, discriminatory, or unlawful. Use validated methods, meaningful human review, notice, an opportunity to contest, and all safeguards required by law. A human reviewer must have authority and enough information to reject the AI recommendation.

7. Code, commands, agents, and project work

Generated code can contain security vulnerabilities, destructive commands, license conflicts, nonexistent packages, incorrect APIs, embedded secrets, performance defects, or logic that appears to work only on the happy path. Never run unfamiliar commands with elevated privileges. Review the exact target of file deletion, migration, deployment, billing, cloud, database, access-control, or network actions. Use backups, version control, least privilege, test environments, and explicit approvals for irreversible operations.

Before production use, run relevant unit, integration, end-to-end, accessibility, security, load, and recovery tests. Inspect dependencies and licenses, pin compatible versions, scan for credentials, and review data flows. Generated tests may repeat the same mistaken assumption as generated code. Security-sensitive and cryptographic code needs expert review. A successful compilation or demo does not prove safety, correctness, scalability, privacy, or compliance.

Agentic tools may take actions through connected systems. Scope permissions narrowly, preview proposed changes, distinguish read-only from write access, and require human confirmation for payments, publication, deletion, account changes, production deployments, external messages, or legal commitments. Do not assume an agent remembers prior constraints across sessions. Monitor long-running tasks and verify the final state in the authoritative system.

8. Creative work, ownership, and authenticity

Generated writing, images, audio, names, designs, and code may resemble existing material or include protected characters, brands, styles, data, or expressions. ZexirAI does not provide a copyright, trademark, patent, publicity-right, or license clearance. Search for conflicts, check applicable licenses, obtain releases, and consult qualified counsel before using important Output commercially. Similar Output may be produced for another user.

Do not use synthetic media to impersonate a person, fabricate evidence, misrepresent endorsements, conceal authorship where disclosure is required, or deceive an audience about an event. Preserve labels or provenance information when law, platform rules, a contract, academic policy, or professional standard requires it. When depicting real people, consider consent, dignity, age, context, and the risk of harassment or reputational harm.

Translations and summaries can change tone, legal effect, or meaning. Compare critical passages with the original and use a qualified translator for legal, medical, immigration, safety, or contractual material. A generated summary is not a substitute for reading the governing document.

9. Bias, fairness, and accessibility

Models learn from data that may reflect historical inequality, cultural assumptions, gaps in representation, offensive language, and measurement error. Output can stereotype people, apply a standard unevenly, or perform worse for certain languages, dialects, disabilities, skin tones, names, or communities. Safety filters can also over-block legitimate discussion or under-block harmful content.

For decisions or content affecting people, examine whether inputs are relevant, whether a proxy stands in for a protected trait, whether error rates differ across groups, and whether affected people can understand and challenge the result. Do not ask a model to infer race, ethnicity, religion, health, disability, sexuality, political belief, immigration status, or another sensitive trait from a name, image, voice, location, or behavior for consequential decision-making.

Generated interfaces and documents may fail accessibility standards. Test keyboard operation, focus order, contrast, zoom, motion, screen-reader labeling, captions, transcripts, and plain-language comprehension with appropriate tools and users. An AI statement that a design is accessible is not an audit or certification.

10. Age-appropriate use

ZexirAI may use an age supplied during onboarding to adjust the language and detail of responses. That personalization is a safeguard, not a guarantee that every answer is suitable, correct, or developmentally appropriate. Users under the age of legal majority should involve a parent, guardian, teacher, or other responsible adult when discussing health, safety, sexuality, self-harm, money, law, relationships, or real-world decisions.

Adults should not direct minors to submit personal contact details, school schedules, precise location, intimate material, government identifiers, payment credentials, or confidential family information. Schools and organizations must conduct their own educational, privacy, accessibility, and safety review, and must not treat generated grades, discipline recommendations, or risk scores as final decisions.

11. Reporting an error or harmful output

If Output appears false, dangerous, biased, infringing, or inconsistent with these policies, stop using it and report the issue. Include the displayed model, approximate time, relevant prompt and output, whether search or attachments were used, and why the issue matters. Remove personal information and secrets that are not needed to reproduce the problem. A screenshot can help when formatting or citations are involved.

We may review submitted material to reproduce the behavior, assess severity, adjust prompts or routing, contact a provider, improve product safeguards, or enforce policies. Reporting does not guarantee a particular model response or immediate correction across all providers. If the issue concerns an imminent threat, account compromise, or active illegal conduct, also contact the appropriate emergency service, platform, or authority rather than relying on a product report.

12. Responsibility and updates

You are responsible for the final decision to use, publish, deploy, send, or act on Output. ZexirAI is responsible for operating the Service as described in the Terms and applicable law, but it does not assume your professional duties or the risks of unreviewed reliance. If your use requires a guaranteed accuracy level, regulated validation, audit rights, or specific service commitment, do not rely on the consumer Service unless a separate signed agreement supplies those terms.

We may update this policy as models, tools, laws, and safety practices change. Material updates will receive notice when required. Questions and error reports may be sent to mail@zexiry.com. Do not include passwords, one-time codes, full payment-card details, or unnecessary sensitive records.

1. Model Limitations: detailed guidance

This section explains how the policy applies to model limitations. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on model limitations, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to model limitations. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns model limitations, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of model limitations as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

2. Source Verification: detailed guidance

This section explains how the policy applies to source verification. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on source verification, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to source verification. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns source verification, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of source verification as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

3. Search Freshness: detailed guidance

This section explains how the policy applies to search freshness. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on search freshness, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to search freshness. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns search freshness, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of search freshness as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

4. Citation Review: detailed guidance

This section explains how the policy applies to citation review. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on citation review, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to citation review. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns citation review, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of citation review as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

5. Mathematical Work: detailed guidance

This section explains how the policy applies to mathematical work. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on mathematical work, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to mathematical work. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns mathematical work, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of mathematical work as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

6. Software Code: detailed guidance

This section explains how the policy applies to software code. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on software code, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to software code. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns software code, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of software code as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

7. Security Advice: detailed guidance

This section explains how the policy applies to security advice. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on security advice, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to security advice. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns security advice, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of security advice as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

8. Medical Information: detailed guidance

This section explains how the policy applies to medical information. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on medical information, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to medical information. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns medical information, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of medical information as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

9. Legal Information: detailed guidance

This section explains how the policy applies to legal information. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on legal information, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to legal information. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns legal information, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of legal information as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

10. Financial Information: detailed guidance

This section explains how the policy applies to financial information. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on financial information, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to financial information. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns financial information, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of financial information as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

11. News And Current Events: detailed guidance

This section explains how the policy applies to news and current events. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on news and current events, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to news and current events. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns news and current events, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of news and current events as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

12. Translations: detailed guidance

This section explains how the policy applies to translations. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on translations, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to translations. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns translations, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of translations as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

13. Summaries: detailed guidance

This section explains how the policy applies to summaries. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on summaries, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to summaries. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns summaries, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of summaries as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.

14. Image Analysis: detailed guidance

This section explains how the policy applies to image analysis. The governing principle is that a user should understand the feature, use only the access and information reasonably needed for the task, and keep qualified human judgment in control whenever an action may affect another person. Product labels, examples, interface text, and automated notices illustrate ordinary operation; they do not replace the specific facts, laws, contracts, professional duties, or provider terms that apply in a particular situation.

Before relying on image analysis, review the requested purpose, the sensitivity of the information involved, the people who may be affected, and the consequences of an error. Use the least sensitive input that can complete the task. Confirm names, dates, amounts, permissions, sources, and destinations before sending, publishing, purchasing, deleting, or connecting anything. If a result appears incomplete, inconsistent, unusually confident, or outside the expected context, pause and verify it through an independent and authoritative source.

ZexirAI may apply technical limits, safety checks, rate controls, account verification, logging, provider routing, and human review to support this policy in relation to image analysis. These measures reduce risk but cannot identify every mistake or misuse. A control that does not block an action is not approval, legal advice, a guarantee of accuracy, or a waiver. Users remain responsible for instructions they provide, content they choose to keep, actions they authorize, and the final use of any output.

When an exception, dispute, or unexpected result concerns image analysis, preserve the minimum information needed to understand what happened: the approximate time, affected account, feature used, relevant settings, safe transaction or request identifiers, and a concise description of the expected and actual behavior. Remove passwords, one-time codes, full payment details, unrelated personal records, and confidential material. Contact support promptly when delay could increase harm, and use emergency or official reporting channels when the situation requires them.

We review the operation of image analysis as the Service, models, providers, security threats, and legal requirements change. We may clarify instructions, adjust safeguards, restrict a capability, or request additional verification. Material policy changes receive notice when required. Existing non-waivable rights continue to apply. If this detailed guidance conflicts with a more specific section of this policy, the more specific section controls for that issue; the Terms and applicable law control the overall relationship.