ASKNOEMA ARTIFICIAL INTELLIGENCE POLICY

How Ajir Research Labs governs, discloses, and limits the use of artificial intelligence in AskNoema.

Effective: on the public launch of AskNoema. Last updated: 24 July 2026. Review cycle: bi-annual.

1. Preamble and Scope

Ajir Research Labs operates AskNoema, a social-intelligence platform that helps people reason through short-form media at the moment of consumption. AskNoema’s reasoning engine is called Episteme. This AI Policy sets out the principles, disclosure standards, operational rules, and limits that govern all AI analysis on the platform. It binds our product engineering, our user-facing communications, and any operators who build on AskNoema. It should be read together with our Terms of Service and Privacy Notice.

2. Definitions

• AskNoema — the product and platform.
• Episteme — the AI reasoning engine that performs analysis within AskNoema.
• AI output — any response, annotation, label, or assessment produced by Episteme.
• Veracity assessment — Episteme’s structured, evidence-based evaluation of a claim in analysed media.
• Operator — a third party that builds a product or service on top of AskNoema.
• Fully AI-generated content — media produced wholly (100%) by an AI system, as distinct from human content that is merely AI-assisted.

3. Foundational AI Principles

Episteme’s development and deployment are governed by the following principles:

• Human dignity and epistemic agency — AI must expand human reasoning, not replace it. Episteme is a reasoning partner that prompts further human inquiry.
• Truth as a standard — the platform is designed never to knowingly facilitate misinformation, and to surface uncertainty, evidence, and contested claims regardless of ideological valence.
• Transparency — you have the right to know when you are interacting with an AI system. AI disclosure is structural, not optional.
• Epistemic humility and honesty — Episteme represents the state of the evidence, including its own uncertainty and limits.
• Non-manipulation — reasoning is presented in a balanced, non-coercive way that preserves your autonomy.
• Responsible scaling — new AI capabilities launch only after safety and ethics review.
• The common good — product and revenue decisions must not be structured to amplify outrage, controversy, or misinformation, even indirectly.

4. AI Transparency and Mandatory Disclosure

AI transparency is a non-negotiable requirement:

• Every AI output carries a disclaimer at the point of first display, stating in substance: “This is an AI-generated response and may contain mistakes. Verify important information independently.”
• You are always clearly informed when you are interacting with Episteme rather than a human.
• Where AskNoema analyses content it assesses to be AI-generated, it displays a clear “AI Generation Suspected” label alongside the supporting evidence.
• Operators building on AskNoema must pass these disclosures through to their end users and may not remove, hide, or dilute them.

This aligns with Article 50 of the EU AI Act (transparency obligations, applicable from 2 August 2026) and Kenya’s proposed AI Bill, 2026.

5. Risk Classification under the EU AI Act

We classify AskNoema’s functions under the EU AI Act (Regulation 2024/1689) and build toward pre-compliance ahead of the 2 August 2026 application date, recognising the regulation’s extraterritorial reach:

• Veracity assessment and AI fact-check responses — Limited Risk (Article 50): mandatory AI disclosure, transparency labelling, and confidence scoring on all outputs.
• Community discourse facilitation — Minimal Risk: voluntary code of practice and moderation transparency.
• General-purpose reasoning via Episteme — general-purpose AI obligations: technical documentation, transparency, copyright compliance, and adversarial testing.
• Biometric or emotion detection, and automated content-removal decisions — High Risk, and not currently deployed. These would require human oversight, conformity assessment, a review-and-appeal mechanism, and user notification before any future deployment.

6. Prohibited Uses of Artificial Intelligence

The following are absolutely prohibited within AskNoema and cannot be unlocked by any operator, partner, or configuration:

• Generating synthetic misinformation or fabricated claims presented as verified truth.
• Producing deepfakes, manipulated audio, or synthetic media of identifiable individuals without clear AI-disclosure labelling.
• Using AI to impersonate journalists, public officials, academic researchers, or institutional sources.
• Deploying emotion recognition or biometric surveillance on users without explicit, informed, and revocable consent.
• Profiling users based on political opinion, religion, ethnicity, health status, or sexual orientation.
• Using AI outputs to influence electoral processes, referenda, or civic voting behaviour.
• Generating content that exploits, demeans, or targets minors.
• Deploying autonomous content moderation that removes user content without a human-review pathway.

7. No Fully AI-Generated Content

AskNoema takes a deliberate and, we recognise, controversial stance: fully (100%) AI-generated content may not be posted publicly on the platform. This is a foundational, zero-tolerance rule — and one of the core reasons AskNoema was created. Human content that is merely AI-assisted remains permitted; only wholly AI-generated media is barred.

Where content is detected as fully AI-generated, it is withheld from public visibility. A creator whose content is flagged may contest the assessment and request human review. While a contest is pending, the content remains withheld from the public until the review is resolved:

• if the flag is overturned on human review, the content is restored;
• if the assessment is upheld, or the creator does not contest it, the content is not published and is removed from our databases.

8. Misinformation and Veracity Assessment

Episteme assesses claims in analysed media and responds proportionately to the type of problem, without removing anyone’s content:

• Factual error (verifiably false) — evidence-based correction with links to primary sources.
• Misleading framing (accurate but misleading context) — contextual information and alternative interpretations.
• Unverifiable claim — labelled unverified, with evidence gaps shown and further inquiry prompted.
• Satire or fiction — genre label only; no correction applied.
• Synthetic / AI-generated content — mandatory AI-disclosure label, confidence scoring, and source tracing.
• Deepfake / manipulated media — high-priority flag, refusal to amplify, and user notification.
• Malinformation (genuine information weaponised out of context) — contextual correction restoring original framing and sourcing.

Every veracity assessment follows a five-stage protocol: claim extraction → source mapping → evidence weighing → confidence scoring (High / Medium / Low / Unverifiable) → output formatting with citations, a confidence score, and an explicit statement of what Episteme cannot verify. Episteme does not decide what is true; it returns the best available evidence map at the time of analysis and invites you to examine primary sources yourself. These are design commitments describing how the system is intended to operate, not guarantees of correctness.

9. Limits on Intervention and Freedom of Expression

AskNoema’s analytical powers are deliberately bounded:

• We do not remove content from external platforms; we provide analysis and context overlays only.
• We do not penalise users for sharing content later found to be false, absent evidence of deliberate bad faith.
• We do not adjudicate contested scientific questions where expert consensus is genuinely divided.
• We respect the constitutional rights to freedom of expression and media (Constitution of Kenya, Articles 33 and 34) and will not use misinformation labels as instruments of political or ideological suppression.

We also note the legal position on criminal “false information” offences: on 6 March 2026 the Court of Appeal declared Sections 22 and 23 of the Computer Misuse and Cybercrimes Act, 2018 unconstitutional; that decision is currently under appeal at the Supreme Court. Pending final determination, AskNoema does not rely on criminal misinformation provisions as a basis for any content action.

10. Human Oversight and Accountability

No Episteme output is the sole determinant of content removal, user sanctions, or institutional-level claims about individuals or organisations. All consequential decisions include a human-review stage. Ajir Research Labs retains the right and the responsibility to monitor, audit, override, and intervene in any AI output. AI remains a tool under human governance; we will not deploy AI systems that operate beyond our ability to monitor and correct.

11. Contesting AI Outputs and Epistemic Engagement

You may challenge any Episteme output. You can contest it publicly, by replying through prompts, or privately, through our feedback system. AskNoema actively promotes this: disagreement with an AI analysis, disagreement with another user’s response, or support for either, are all welcomed as healthy epistemic engagement. Contesting a consequential assessment also triggers the human-review pathway in section 10.

12. Session Privacy and the AI Bank

Your AI interactions are private by default. Your prompts and Episteme’s responses are visible only to you unless you choose otherwise. You may opt to publish selected AI interactions to the AskNoema AI Bank, a public repository of shared interactions. Sharing is entirely user-initiated — nothing from your session is made public without your action. How we store and protect this data is set out in our Privacy Notice.

13. Fairness and Non-Discrimination

Episteme’s analytical frameworks are tested and continuously monitored for bias across:

• Geographic / regional — training and validation data must include diverse African, Global South, and non-Western sources; Kenyan and Nairobi-relevant contexts must be expressly represented.
• Linguistic — Swahili, Sheng, and other Kenyan vernaculars must be handled with the same analytical care as English.
• Political / ideological — the same veracity standards apply regardless of political alignment; asymmetric treatment is a material breach of this policy.
• Source authority — independent, community-based, or non-institutional sources are not systematically penalised in confidence scoring.
• Temporal — assessments reflect evidence as of the analysis date, not stale training data, and disclose recency limits.

14. Intellectual Property and Creator Content

• Our analysis constitutes commentary and reference — consistent with fair-use and fair-dealing principles — not reproduction or redistribution of third-party content.
• We do not store, cache, or redistribute original creator content beyond what is technically necessary for real-time analysis.
• Creators and rights-holders who object to analysis of their content are provided a documented opt-out mechanism.
• We do not use creator content to train or fine-tune AI models without explicit, revocable consent.

15. Third-Party AI Providers

Episteme’s reasoning may be powered in part by third-party AI providers (AI companies) engaged as service providers under written agreements. These are processing arrangements, not content or data-sharing partnerships: providers process the relevant content only transiently, and only to return the analysis you requested. We do not sell your data, and we do not hand it to any provider for that provider’s own purposes. Where processing involves providers outside Kenya, we apply the cross-border safeguards described in our Privacy Notice.

16. Operators and Partnerships

• Operators integrating AskNoema must preserve every AI-disclosure obligation for their end users and may not suppress or obscure it.
• Operators are bound by the prohibited uses in section 6 and may not use the platform for mass surveillance, behavioural targeting, or political propaganda.
• Institutional and educational partnerships must incorporate AI-literacy components, with heightened protection for any student data.
• Partnership enquiries: partnerships@ajirresearch.com.

17. Children and Vulnerable Users

AskNoema is available to users aged 15 and above; users aged 15–17 require verifiable parental or guardian consent, and users under 15 may not use the platform. For younger and vulnerable users, we apply age-appropriate, non-manipulative design and heightened safeguards, consistent with Kenya’s Data Protection Act, 2019 (s. 33), the ODPC’s guidance on children’s data, and the protections in Kenya’s Artificial Intelligence and Other Emerging Technologies Policy, 2026.

18. Alignment with Law and Governance

This policy is designed to align with the EU AI Act (Articles 50 and general-purpose AI obligations), the Kenya Artificial Intelligence and Other Emerging Technologies Policy, 2026, the Data Protection Act, 2019 and the oversight of the ODPC, and the Constitution of Kenya (including Articles 33 and 34). Where obligations differ across jurisdictions, we apply the standard most protective of users and their rights.

19. Changes and Review

We review this policy on a bi-annual cycle and may update it to reflect new laws, regulators’ guidance, court decisions, or features. When we make material changes, we revise the “Last updated” date and notify users before the changes take effect.

20. Contact

Questions about this policy can be directed to support@ajirresearch.com. Suspected security or safety issues can be reported to security@ajirresearch.com.