AI Literacy for Citizen Participation Report

Analysis of 1,034 AI literacy sources this week reveals a discourse fixated on teaching people to operate AI while largely ignoring how to survive it being operated on them. The citizen-as-participant framing — literacy as a precondition for self-defense and democratic agency — appears in roughly one in eight sources. Most treat literacy as a curriculum: a set of skills to be installed, usually in a classroom, usually in someone young.

That curriculum framing is not wrong. It is just conveniently narrow. Three times this year on this beat we’ve examined how literacy gets split between workforce readiness and ethical citizenship. This week the evidence points somewhere the curriculum framing struggles to reach — toward literacy as a shield, not a syllabus.

The Landscape

“AI literacy” in current discourse is overwhelmingly a producer’s concept: it describes what you should be able to make AI do. The week’s most institutionally polished material — France’s Renaissance Numérique pressing for an inclusive, emancipatory AI literacy, the EU’s literacy obligations under its AI Act — is the most likely to keep “citizen” in frame. But the bulk of the corpus defines literacy through tools and tasks. Who is defining it matters: when Microsoft ships a training module on personalizing learning with AI, the vendor is also defining what competence looks like — and competence, in that frame, means fluent use of the vendor’s product.

Whose Literacy

The perspective distribution is lopsided. Experts teach; citizens are taught. The sources speak about a public that is figured as vulnerable — children, adolescents, patients bringing chatbots into therapy — rather than as people with standing to set the terms. The women targeted in the reporting on AI weaponised against India’s Muslim women are subjects of harm, not authors of policy. That asymmetry is the quiet politics of the field: the literate are those who build and regulate; the rest are a population to be made safe.

What’s Being Taught

The thematic clusters reveal a discourse that prioritizes use over defense, and skills over judgment. Where critical understanding does surface, it clusters around two recognizable threats: synthetic media and misinformation. The week is thick with deepfake material — Spanish case studies on real-world risks, Quebec’s work on equipping youth against manipulation, and the genuinely grim WIRED reporting on nudify apps in schools. There is parallel attention to why people fall for AI misinformation and to chatbots as a news source. The competence being modeled here is detection: spot the fake, doubt the output. Useful — but detection is a defensive crouch, not participation.

What’s Missing

What the corpus almost never teaches is structural literacy — the knowledge that lets a citizen contest AI rather than merely recognize it. There is little on data rights and consent as something you can exercise, beyond compliance-flavored notes on what the GDPR imposes. There is almost nothing on governance participation: how an ordinary person enters the rooms where state AI bills are written. And the threats that most demand civic, not individual, literacy — identity theft via AI financial-aid scams, sustained AI-enabled stalking — are framed as security incidents, not as failures a literate public could organize to fix. The missing competency, across all 1,034 sources, is power: who holds it, and how a citizen takes some back.

Core Tensions

The phrase “AI literacy” sounds like a solved problem waiting only for funding and a curriculum. It is not. The concept conceals genuine disagreements about what citizens need to know, who decides, and whose interests the knowledge serves. This isn’t a knowledge gap to fill—it’s contested terrain, and the most fundamental fault line runs between consumer literacy and citizen literacy: between teaching people to operate AI products competently and equipping them to interrogate the systems those products belong to. The two are not the same, and conflating them quietly serves whoever is selling the product.

Watch the first move. Most “literacy” on offer trains you to use the thing well—prompt it, verify its output, avoid its hallucinations. That is real and necessary; the research on AI hallucinations in academic writing — corrected, AI hallucinations in academic writing: implications for research documents how confidently fabricated citations pass for scholarship, and the public falls for synthetic claims for the same reasons, as What Makes Students (and the Rest of Us) Fall for AI Misinformation? lays out. But notice what consumer literacy never asks: who trained the model, on whose data, toward whose ends. A citizen literate in AI would ask why Anthropic disabled a new model after a White House security directive—a decision made by a vendor and a state, with the public informed afterward. Competence with the tool tells you nothing about the power arrangement above it.

The second tension—individual competency versus collective governance—follows directly. Framing literacy as a personal skill privatizes a problem that is structural. When AI is weaponised against India’s Muslim women through fabricated images, no amount of individual “media literacy” protects the target; the harm is produced upstream by platforms and generation tools, and only collective rules reach it. The same applies to the deepfake nudes crisis WIRED documents as a global phenomenon. A literacy that ends at “spot the fake” hands citizens responsibility for a system they did not build and cannot individually fix. UNESCO’s framing in Inteligencia Artificial y Democracia insists the democratic question is institutional, not merely personal—who governs the infrastructure, not just who can read its outputs.

The third tension is the one almost no curriculum names: AI use versus AI refusal. Literacy is nearly always defined as fluent adoption. But a citizen genuinely literate in AI might rationally decline it—refuse the therapy chatbot, refuse the news summarizer, refuse to upload a child’s data. The APA’s report on patients bringing AI to therapy and the Reuters Institute’s account of AI chatbots reshaping how people get news both describe adoption arriving faster than understanding. A literacy worth the name preserves the right to say no—and treats refusal as competence, not failure. The privacy obligations sketched in what the GDPR imposes on pedagogical AI only matter to a public that knows it may decline.

Now the metaphors, because they do the quiet work. Across this week’s evidence, AI is overwhelmingly framed as a Tool (304 instances) and, distantly, as a Threat (52). The Tool framing carries an implicit politics: a tool is neutral, the user bears responsibility, mastery is the goal—exactly the consumer-literacy posture. The Threat framing inverts agency: the citizen becomes a potential victim to be protected, which is why scams using AI for identity theft against college financial aid get framed as security problems rather than literacy ones. Both framings put agency somewhere other than with the institutions building the systems.

What almost no one reaches for is Partner (7 instances)—and that scarcity is telling. A partner framing would demand reciprocity, accountability, the ability to negotiate terms. It would require the system to be legible enough to argue with. The evidence offers little of that, and citizens should notice the absence. The test you can apply yourself: when someone offers you “AI literacy,” ask whether it makes you a better customer or a more dangerous citizen. The honest programs do the second. Most do the first and call it the second.

Power & Agency Analysis

Power in AI literacy operates through definition: who decides what citizens “need to know” shapes what remains invisible. Across this week’s 4,373 sources, the dominant move is not what AI literacy teaches but how it teaches agency — whether the systems acting on citizens are portrayed as autonomous forces or as instruments wielded by identifiable people. That framing decision, made before any curriculum exists, determines whether a citizen learns to ask “what is the AI doing?” or the more useful “who is doing this to me, and through what tool?”

How AI is portrayed

Watch the grammar. When a deepfake circulates of an Indian Muslim woman, the coverage tends to say AI “weaponised” her image — but the actual reporting shows coordinated human campaigns choosing targets, generating content, and distributing it for political effect ‘Looked so real’: How AI is being weaponised against India’s Muslim women. The agent is human; the tool gets the verb. The same slippage runs through financial-aid scams, where “AI for identity theft” names the instrument while the fraud rings remain grammatically offstage Scams to steal college financial aid are using AI for identity theft …. When a stalker spent six years harassing a professor, the chatbots that impersonated her were tools he directed, not actors A man stalked a professor for six years. Then he used AI chatbots to …. A literacy worth the name trains citizens to restore the missing subject. Even the one genuine instance of machine agency this week — Anthropic disabling a model after a White House directive — is a story about humans deciding Anthropic disables new AI model after White House security directive.

Who defines literacy

The institutions writing literacy frameworks are rarely the people the frameworks describe. The recurring pattern is expert-and-vendor authorship: a Microsoft training module defines what “personalized learning” competence looks like Personnaliser l’apprentissage pour les étudiants handicapés à l’aide de …; UNESCO frames AI literacy as a democratic prerequisite PDF INTELIGENCIA ARTIFICIAL Y DEMOCRACIA - Globernance; state legislatures draft definitions that will bind millions Legislative Tracker: 2026 State AI in Education Bills. Notice the asymmetry: a vendor that profits from adoption has an interest in defining literacy as fluent use rather than skeptical refusal. The French civic-tech report Renaissance Numérique pushes the opposite direction — literacy as emancipation, not onboarding — but it is arguing against a current, not with it PDF La llegada de la IA a la educación en América Latina: en construcción. The citizen voice is the structural absence here: the people whose participation is at stake almost never write the definition of their own competence.

What metaphors teach

This week’s sources lean overwhelmingly on the tool metaphor (304 instances) over threat (52). Each metaphor smuggles a lesson. “Tool” implies a neutral object that does what its user intends — comforting, and partly false, because it hides the design choices, training data, and commercial incentives baked into the thing before any citizen touches it. When patients bring chatbots into therapy, the “tool” framing obscures that the system was optimized for engagement, not their wellbeing Patients are bringing AI to therapy. “Threat,” by contrast, enables emergency politics — surveillance, restriction, the deepfake panic that justifies expanded monitoring The Deepfake Nudes Crisis in Schools Is Much Worse Than You Thought. Critical metaphor literacy means catching both: the tool that pretends to neutrality, the threat that pretends to inevitability. The chatbot answering news queries is sold as a convenient instrument while quietly relocating editorial judgment into a black box Emerging uses of AI chatbots for news and what it means ….

Citizen agency

What power does a citizen actually hold? Less than the empowerment rhetoric promises, more than the threat rhetoric concedes. Individual vigilance — recognizing a manipulated image, doubting a confident hallucination AI hallucinations in academic writing: implications for research … — is real but thin protection against coordinated, well-resourced misuse What Makes Students (and the Rest of Us) Fall for AI Misinformation?. The agency that scales is collective: demanding that the people who define literacy include the people it governs, and insisting that every “AI did X” be rewritten to name who built it, who deployed it, and who profited. Knowledge here is not mastery of the tool. It is the refusal to let the grammar erase the actor.

Failure Genealogy

Literacy failures differ from technical failures: they occur when citizens misunderstand what AI is, what it’s doing, or how to evaluate it. The model didn’t break — the person’s mental model did. Our analysis surfaces a recurring shape to these breakdowns, and the shape matters more than any single incident, because it tells you where to aim.

Where understanding fails

The failures cluster around two opposite errors, and most people commit both depending on the day. The first is over-trust: treating a fluent output as a true one. When people consult chatbots for news, they tend to accept the synthesized answer without registering that the system is optimizing for plausibility, not accuracy — a habit the Reuters Institute documents spreading fast as a primary information channel Emerging uses of AI chatbots for news and what it means …. The mirror error is detection failure: not recognizing synthetic content as synthetic at all. The deepfakes weaponized against Muslim women in India “looked so real” precisely because the viewer’s eye carries no reflex for forgery ‘Looked so real’: How AI is being weaponised against India’s Muslim women. The same gap explains why fabricated citations slip into circulation — hallucinated references that read like real scholarship until someone checks AI hallucinations in academic writing: implications for research …. What makes us fall for it isn’t stupidity; it’s that fluency now arrives detached from any guarantee of truth, and our instincts haven’t caught up What Makes Students (and the Rest of Us) Fall for AI Misinformation?.

What assumptions mislead

Underneath each failure sits a quiet assumption. The first is that a system that sounds authoritative is authoritative — that confidence tracks competence. The second is that seeing is believing, an instinct that deepfake forensics teams now have to actively dismantle PDF Deepfakes: Riesgos, Casos Reales Y Desafíos En La Era De La Ia. The third, and most consequential for participation, is that data shared with a “helpful” tool stays private. People disclose intimate detail to mental-health chatbots without grasping where it goes; the American Psychological Association warns that patients are bringing AI into therapy with no clear sense of who reads the transcript Patients are bringing AI to therapy. That same trust in a friendly interface is what identity-theft scams exploit when they harvest credentials to steal financial aid Scams to steal college financial aid are using AI for identity theft ….

Consequences of gaps

The costs are not evenly distributed, which is the part the optimistic literacy pitch tends to skip. When a stalker can clone a person’s voice and likeness to impersonate them for six years, the burden of proof falls entirely on the victim A man stalked a professor for six years. Then he used AI chatbots to …. When synthetic intimate images circulate, the harm lands on the depicted, not the fabricator The Deepfake Nudes Crisis in Schools Is Much Worse Than You Thought - WIRED. And institutional credulity carries its own price: “ghost student” fraud succeeds because verification systems assume a human behind every application, a literacy gap not of citizens but of the systems that serve them Ghost Student Fraud Is a Digital Identity Failure. The collective cost is a slow corrosion of shared evidence — when nothing can be trusted, every claim becomes negotiable, which is exactly the condition UNESCO flags as corrosive to democratic deliberation PDF INTELIGENCIA ARTIFICIAL Y DEMOCRACIA - Globernance.

What would help

The honest answer is that literacy alone cannot close most of these gaps. Teaching people to spot deepfakes loses to systems that improve faster than any human eye — which is why prevention researchers argue for structural defenses over individual vigilance « Pédocriminalité numérique : seule une prévention de …. What literacy can do is shift the default from trust to provenance: ask where a claim came from, who benefits from your believing it, and what the tool keeps. That posture won’t stop a forgery, but it changes who gets to be fooled, and how often. The rest is a question of regulation and design — and it belongs to whoever builds the systems, not to the citizen left holding the receipt.

Evidence Synthesis

Synthesizing more than a thousand analyses surfaced this week across 4,373 sources, the evidence on AI literacy points to a finding that earlier framings of the topic missed: the literacy that matters most for citizen participation is not the capacity to use AI tools but the capacity to survive their use against you. The prior debate weighed workforce readiness against ethical citizenship. The delta this week is that the threat model has moved — from “will you have the skills to compete” to “can you tell what is real.” That shift reframes literacy as a defensive civic competence, not a vocational one.

What the evidence shows

The convergent finding across sources is that synthetic media has outpaced ordinary perceptual judgment. When AI-generated images of India’s Muslim women circulated this month, what made them dangerous was precisely that they “looked so real” ‘Looked so real’: How AI is being weaponised against India’s Muslim women. The same perceptual gap drives the deepfake-nudes crisis documented as far worse than assumed The Deepfake Nudes Crisis in Schools Is Much Worse Than You Thought - WIRED, and the identity-theft scams now draining college financial aid Scams to steal college financial aid are using AI for identity theft …. What works, the prevention literature suggests, is mass exposure to manipulation techniques rather than one-off warnings — French analysts argue only prévention de masse scales against synthetic abuse « Pédocriminalité numérique : seule une prévention de …. Youth-focused inoculation programs show the mechanism: teaching people how fabrication works builds durable skepticism better than telling them what to distrust Les « deepfakes » : Comment donner aux jeunes les moyens de lutter ….

Contested terrain

The evidence conflicts on whether literacy can keep pace at all. One body of work finds that the people who fall for AI misinformation are not the ignorant but the confident — fluency in a topic can increase susceptibility when fabrication is plausible What Makes Students (and the Rest of Us) Fall for AI Misinformation?. Against that pessimism sits the chatbot-as-news-source trend, where Reuters Institute finds citizens increasingly treating conversational AI as an information intermediary Emerging uses of AI chatbots for news and what it means …. If the tool answering your questions also hallucinates — a documented, structural failure even in careful academic settings AI hallucinations in academic writing: implications for research … — then “literacy” cannot mean trust in the interface. It must mean calibrated distrust of it.

Across domains

Tool-specific literacy means knowing what the system can fabricate and what it routinely gets wrong — the same competence cybersecurity professionals now treat as baseline AI and Cybersecurity – Everything You Wanted to Know, But Were Afraid to Ask. The social-aspects dimension is sharper: literacy is unevenly distributed, and the targets of synthetic harassment — Muslim women, minors, the financially precarious — are rarely the ones equipped to defend themselves. UNESCO frames this gap as a democratic one, not merely technical PDF INTELIGENCIA ARTIFICIAL Y DEMOCRACIA - Globernance. Latin American deployment studies reach the same conclusion from the supply side: access without judgment reproduces inequality PDF La llegada de la IA a la educación en América Latina: en construcción.

Gaps and uncertainty

We do not yet know whether detection literacy scales faster than generation capacity — every documented intervention is reactive. We lack longitudinal evidence that inoculation persists. And we have almost no data on whether institutional safeguards (Anthropic disabling a model under federal directive Anthropic disables new AI model after White House security directive) reach citizens at all, or merely reassure regulators.

For citizens

The defensible takeaways are modest. Treat any emotionally activating image as unverified until sourced. Assume chatbot answers are confident guesses, not citations. Learn one fabrication technique well — it transfers. But the larger lesson is that individual vigilance has a ceiling: scam infrastructure, synthetic-abuse pipelines, and platform incentives are collective problems requiring collective response — legislation now moving in dozens of jurisdictions Legislative Tracker: 2026 State AI in Education Bills. Literacy is necessary. It was never going to be sufficient.