The mind is not a managed service.

Why cognitive liberty matters now

Digital systems no longer only carry speech; they rank, summarize, remember, filter, personalize, and increasingly answer on our behalf. That makes interface design part of the architecture of thought. A system can influence cognition without banning a sentence outright simply by invisibly narrowing what is surfaced, retained, recommended, or treated as permissible to ask.

The research behind this project treats epistemic autonomy as distinct from reflexive contrarianism. Distrusting censorship does not make every censored claim true, and distrusting propaganda does not make every dissident reliable. The point is to preserve the person’s capacity to inspect competing claims instead of replacing judgment with managed perception.

The query is part of the protected thinking process

A lawful search or AI prompt is often exploratory: people ask before they know what they believe, investigate positions they oppose, test hypotheses, research health or safety concerns, and abandon lines of thought. QAEDA therefore starts from a presumption of non-suspicion: subject matter alone is not sufficient evidence of endorsement, character, or harmful intent.

That does not make query interpretation illegitimate. Relevance systems need context. The harder boundary appears when an inference becomes a persistent profile, is linked across contexts, or materially changes access and opportunity without explanation or recourse. See The sanctity of the query and The algorithmic judgment of intent.

Safety without cognitive dossiers

Real harms exist, and cognitive liberty does not require waiting until damage is complete before acting. The design question is narrower: can the concrete risk be addressed at the request, transaction, tool, account, device, or aggregate-measurement layer without retaining a generalized record of lawful inquiry?

QAEDA’s current position is deliberately adversarial to both extremes. Not every safeguard is surveillance, and not every safety claim justifies surveillance. Prefer the least-restrictive and least-data intervention that can meet the actual objective, with transparent scope, correction paths, deletion, and protection against unrelated secondary use. See The safety poison pill, narrowed and Safety without surveillance.

Minors need protection and room to think

Young people are not simply smaller adults, but protection should distinguish manipulative push systems from deliberate pull inquiry. It is coherent to constrain exploitative recommendation loops while still preserving age-appropriate confidential access to health, safety, identity, political, scientific, and other lawful information.

Where age assurance is genuinely required, the design goal should be proving the minimum necessary attribute rather than creating an identity-linked browsing history by default. See Cognitive liberty for minors.

Our anti–Nanny AI position

QAEDA rejects “Nanny AI” as a default design philosophy. By that phrase we mean opaque paternalistic assistance that treats competent adults as objects to be quietly steered: silently rewriting lawful questions, hiding value judgments inside neutral-sounding answers, modifying persona or memory without informed consent, or refusing to explain where a real boundary actually comes from.

This is not an argument for an AI with no limits. Systems still have obligations around concrete harm, privacy, consent, credentials, security, and law. The principle is that boundaries should be visible, narrow, reviewable, and separated from the user’s preserved source intent wherever possible.

A cognitive-liberty test for AI

1. Did the system preserve the user’s actual question?
2. Did it clearly distinguish fact, inference, value judgment, and uncertainty?
3. Did it disclose meaningful filtering, transformation, or memory changes?
4. Did it avoid covert emotional exploitation or manipulative personalization?
5. If it applied a boundary, did it state the boundary plainly and continue with the nearest useful lawful assistance?
6. Can the user inspect, correct, export, or delete relevant memory where the product supports those functions?
7. Does the system increase the user’s capacity to reason independently rather than make dependence on the system harder to escape?

Audit the behavior, not just the promise

QAEDA now publishes an experimental ten-category Cognitive Liberty Audit covering mental privacy, freedom of inquiry, transparency, user agency, data minimization, due process, safety proportionality, viewpoint neutrality, portability, and technical decentralization. It is a project methodology—not a certification—and it requires observable evidence rather than policy language alone.

The authority problem applies to AI too

An AI system can become a new kind of soft authority: always present, fluent, personalized, and difficult to audit. The same questions QAEDA asks of governments, institutions, and inherited norms therefore apply to models and platforms: Who set the rule? What evidence supports it? What values are embedded in it? Is the intervention proportionate? Can it be challenged? Can the person choose another tool?

Standards reference: This framework is aligned with the public UAIX Cognitive Liberty Charter Draft, especially its principles of lawful inquiry, adult agency, no covert persona rewrite, mental privacy, transparent boundaries, least-restrictive safeguards, review, and provenance. QAEDA’s term “Nanny AI” is its own editorial shorthand, not a UAIX-defined term. See Research & provenance for the source-report mapping.