AI Literacy for Citizen Participation Report

Analysis of 975 AI literacy sources drawn from this week’s 4,688-article corpus reveals a discourse that has quietly changed its subject. Where earlier framings treated literacy as workforce preparation and misinformation defense — the twin goals this publication has traced through a year of coverage — the material now clustering under “AI literacy” is overwhelmingly about harm done to citizens rather than skills built by them. The citizen-as-active-participant framing — someone who might shape AI governance, not merely survive it — appears in fewer than one source in five. Most treat literacy as a defensive crouch: learn to spot the deepfake before it reaches you.

The Landscape

Watch what counts as “literacy” now. The dominant vocabulary is threat-detection. This week’s sources cluster around a manipulated-election story in which an AI system allegedly purged millions of names from voter rolls Election AI Deleted 5.5M Voters While World Governed Deepfakes Instead, around deepfakes entering courtrooms as evidence AI-generated evidence showing up in court alarms judges, and around the French municipal campaigns already bracing for synthetic disinformation Municipales 2026 : IA, deepfakes et désinformation, la démocratie …. This is a real shift in the evidence, not just tone: a year ago the argument was that citizens needed AI fluency to stay employable. Now the framing is that citizens need it to stay unmanipulated. Notice who benefits from that reframing — it moves the burden of defense onto the individual reader while leaving the systems that generate the manipulation untouched.

Whose Literacy

The teaching voices are institutional, and they are not neutral. Election authorities, election-security researchers, and civil-society watchdogs supply most of the framing — the Brennan Center on whether AI fights or fuels disinformation Does AI Fight or Fuel Election Disinformation?, CIVICUS on future-proofing elections PDF Future-Proofing Elections Against Deepfake Disinformation, the European Parliament’s research service on information manipulation PDF Information manipulation in the age of generative artificial intelligence. Vendors write their own literacy, too: OpenAI’s rules on political use tell citizens what the platform will and will not permit Political Campaigning Restrictions — a document that teaches deference, not judgment. The citizen appears almost entirely as the object of instruction, rarely as its author. Renaissance Numérique’s report on deploying AI literacy for democratic life is one of the few that treats the public as a body with agency rather than a surface to be protected PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour une.

What’s Being Taught

The curriculum, such as it is, splits into two piles. The larger pile is protective: recognize cloned voices and synthetic faces Deepfakes 2026: clonación de voz, fraude con IA y protección legal, understand how sextortion now scales through generation tools Sextortion in the Age of AI, grasp how chatbots facilitate grooming Platform-Facilitated Grooming and AI Chatbots. The smaller pile is procedural: how detection actually works, and why it fails — the false-positive problem that turns “AI literacy” into a source of new injustice when detectors wrongly flag human work False Positives in AI Detection: Complete Guide 2026. Almost nowhere is the citizen taught to interrogate the systems themselves — the audit-and-accountability competence that Amnesty’s toolkit gestures toward Algorithmic Accountability Toolkit - Amnesty International.

What’s Missing

The gap is governance. Sources teach citizens to detect and to protect, but scarcely to participate — to file a challenge, demand an audit, or shape the rules under which policing algorithms operate The Dangers of Unregulated AI in Policing - Brennan Center for Justice. The Swiss experiment in bringing AI into democratic deliberation is a rare exception Quand l’IA s’invite dans le débat démocratique suisse. Missing, too, are the people most exposed and least addressed: non-English speakers, low-connectivity communities, and anyone whose defense against a synthetic accusation is not a detector but a lawyer they cannot afford. A literacy that only teaches fear produces compliant subjects, not participating citizens.

Core Tensions

The concept of “AI literacy” conceals genuine tensions about what citizens need to know and why. Our analysis of this week’s 4,688 sources maps at least six live contradictions, but the most fundamental is this: is AI literacy a technical skill you acquire, or a critical stance you take? The two answers pull in opposite directions, and the difference decides who literacy is for. This isn’t a knowledge gap to fill—it’s a contested terrain, and the people funding the curricula have a stake in which definition wins.

Technical skill versus critical understanding. The dominant framing treats literacy as competence: learn the prompts, understand the model, use the tool well. But the evidence keeps surfacing a second definition, one built around suspicion rather than fluency. When the U.S. Secret Service describes the mechanics of AI-driven sextortion Sextortion in the Age of AI, or when the European Parliament catalogues how children are targeted by deepfakes PDF Children and deepfakes - europarl.europa.eu, the “skill” being taught is refusal—the capacity to recognize a synthetic artifact and not act on it. Renaissance Numérique’s October report frames literacy explicitly as democratic infrastructure rather than employability PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour une. Watch this move: the same word covers “use it better” and “distrust it harder,” and vendors overwhelmingly fund the first.

Individual competency versus collective governance. The second tension is scale. A literate citizen who can spot a deepfake is still helpless against 5.5 million voters deleted by an election system operating below the threshold of individual perception Election AI Deleted 5.5M Voters While World Governed Deepfakes Instead. The obsession with teaching individuals to detect fakes—while the actual harms are structural and infrastructural—is a quiet transfer of responsibility. Amnesty International’s accountability toolkit locates the problem in institutions and audits, not in personal vigilance Algorithmic Accountability Toolkit - Amnesty International, and the Brennan Center’s work on policing shows how “literacy” can’t substitute for regulation the individual has no power to demand The Dangers of Unregulated AI in Policing - Brennan Center for Justice. When you’re told the solution is your own education, ask who benefits from the problem staying yours.

Consumer literacy versus citizen literacy. OpenAI’s own political campaigning restrictions define acceptable civic behavior on the platform’s terms Political Campaigning Restrictions—a reminder that the environment where citizens practice “literacy” is privately governed. The Brennan Center’s dual finding, that AI both fights and fuels election disinformation Does AI Fight or Fuel Election Disinformation?, captures the trap: a consumer learns to navigate the product; a citizen needs to interrogate the product’s role in the public sphere. Detection tools that falsely accuse real people False Positives in AI Detection: Complete Guide 2026 show what happens when consumer-grade “literacy” gets weaponized as judgment—the literate act becomes accusation, and the accused have no appeal.

The metaphors doing the work. Across the corpus, AI is overwhelmingly framed as a Tool (304 instances) and far less as a Threat (52). The tool metaphor is not neutral: it presumes a competent user in control, which quietly loads the entire burden of outcomes onto the person holding it. If AI is a tool, then misuse is your fault—a framing that dovetails suspiciously well with the individual-competency model above. The Threat framing, by contrast, implies defense and regulation, agency held collectively. What almost no one reaches for is the Partner metaphor (7 instances)—and its scarcity is telling, because a partner is something you negotiate with, hold accountable, and can refuse. The Berkeley agentic-AI summit surfaced exactly this vacuum: you cannot send an AI agent to prison Impossible d’envoyer un agent IA en prison. Accountability requires an agent that can be held responsible; the tool metaphor guarantees there isn’t one.

Citizens can test any literacy program with a single question: does it teach me to use the tool, or to judge the system? The European Parliament’s work on generative manipulation PDF Information manipulation in the age of generative artificial intelligence and CIVICUS’s election research PDF Future-Proofing Elections Against Deepfake Disinformation both point the same direction: the literacy that matters is the one that survives contact with power you don’t hold.

Power & Agency Analysis

Power in AI literacy operates through definition: whoever decides what citizens “need to know” also decides what stays invisible. Our analysis of this week’s 4,688 sources finds the language of the field tilted heavily toward one framing — the “tool” metaphor appears 304 times against just 52 invocations of AI as “threat.” That six-to-one ratio is not neutral vocabulary. It is a settled position on where agency lives, and it teaches citizens to look in the wrong place.

How AI Is Portrayed

Watch the grammar of any headline about AI harm. “Election AI Deleted 5.5M Voters” reads the story of an actual purge Election AI Deleted 5.5M Voters While World Governed Deepfakes Instead — the software is the subject, the verb is “deleted,” and the humans who procured, configured, and deployed that system have vanished from the sentence. This is the dominant move: agency floats up to the machine. When a Berkeley summit on agentic AI concluded that you cannot send an AI agent to prison Impossible d’envoyer un agent IA en prison, it named the exact accountability vacuum the “AI decides” framing creates. The same slippage runs through predictive policing, where automated outputs are treated as findings rather than choices The Dangers of Unregulated AI in Policing, and through military targeting, where systems turn “could” into “must” and quietly relocate the decision from a commander to a query The queryable war: When AI turns ‘could’ into ‘must’. For a citizen, the lesson embedded in this grammar is disempowering: the machine acted, so no one is answerable. The literacy that matters most is the reflex to re-insert the missing human — the vendor, the agency, the official who pressed deploy.

Who Defines Literacy

The frameworks telling you what to learn are themselves authored by interested parties. Renaissance Numérique’s October 2025 report positions civic AI literacy as a democratic project PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour une — but most operative definitions arrive from the platforms themselves. When OpenAI publishes rules on what its models may do in campaigns Political Campaigning Restrictions, it is drafting the boundaries of civic conduct as a private terms-of-service document. Gartner, meanwhile, frames the coming years as a “battle for authority” and sells readiness for the aftermath Gartner has confirmed the battle for authority is happening right now. Notice who is absent: the citizen appears as a subject to be prepared, rarely as a voice defining what preparation should mean.

What Metaphors Teach

The “tool” metaphor’s 304 appearances flatter the user — a tool is inert, obedient, yours. That framing obscures dependence, data extraction, and the fact that the tool has commercial goals of its own. It also disarms suspicion at exactly the wrong moment: AI detection systems marketed as neutral instruments generate false positives that damage real people False Positives in AI Detection: Complete Guide 2026, and AI-generated evidence now enters courtrooms wearing the credibility a “tool” is assumed to carry AI-generated evidence showing up in court alarms judges. The 52 “threat” invocations do the opposite work — they externalize danger onto the technology, which conveniently exonerates the humans running sextortion and grooming operations through it Sextortion in the Age of AI. Critical metaphor literacy means holding both against the evidence: the tool that manipulates and the threat that someone built and profits from PDF Information manipulation in the age of generative artificial intelligence.

Citizen Agency

The honest answer is that individual agency is limited, and Amnesty International’s accountability toolkit says so plainly — the leverage is collective, structural, aimed at the institutions that deploy these systems Algorithmic Accountability Toolkit. What a citizen can do alone is refuse the grammar that erases responsibility. The Brennan Center’s finding that AI both fights and fuels disinformation depending on who wields it Does AI Fight or Fuel Election Disinformation? is the whole lesson: the variable is never the machine, it is the hand behind it. Knowledge here is not defense against the technology. It is the capacity to name the actor — and to demand that the courts, agencies, and platforms currently hiding behind “the AI did it” answer as the humans they are.

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. Our analysis across this week’s 4,688 sources documents a recurring shape to these breakdowns — not exotic edge cases, but ordinary moments where a reasonable person, lacking a specific piece of understanding, gets played.

Where understanding fails

The failures cluster around two opposite errors, and both are dangerous. The first is over-trust: treating an AI output as authoritative because it is fluent, formatted, and instantaneous. The Berkeley agentic-AI summit surfaced a version of this that ought to alarm anyone — the observation that you cannot send an AI agent to prison, meaning accountability evaporates precisely where confidence is highest Impossible d’envoyer un agent IA en prison: les citations …. The second error is the mirror image: reflexive under-trust that discards good information as “probably fake,” which is exactly the corrosion the disinformation researchers warn about — a public so primed to doubt that genuine evidence loses its force Does AI Fight or Fuel Election Disinformation?.

Detection is where the gap is widest. Citizens routinely believe they can spot a deepfake by eye. They cannot. Voice-cloning fraud now works on a few seconds of audio Deepfakes 2026: clonación de voz, fraude con IA y protección legal, and even institutional detectors misfire — the “AI-detection” tools sold as a fix produce false positives at rates that ruin real people False Positives in AI Detection: Complete Guide 2026. The literacy failure here is the assumption that detection is a solved problem someone else is handling.

What assumptions mislead

Three assumptions do most of the damage. First: seeing is believing — the deep intuition that a video or a voice is evidence of the event it depicts. That assumption is now false by default, and courts are feeling it, with judges alarmed by AI-generated material entering as evidence AI-generated evidence showing up in court alarms judges. Second: the platform is protecting me — the belief that some backstop filters manipulation before it reaches you, when safeguards have measurably eroded Misinformation may get worse in 2024 election as safeguards erode | AP News. Third, and quietest: my data is not the product — the assumption that a friendly conversational interface is not also an extraction machine. Grooming research on AI chatbots shows how the intimacy of the interface is itself the vector Platform-Facilitated Grooming and AI Chatbots.

Consequences of gaps

The costs are not evenly distributed, and that is the point worth naming. Sextortion enabled by generative tools falls hardest on the young and isolated, who are least equipped to recognize the manipulation and most ashamed to report it Sextortion in the Age of AI; children are a documented target class for synthetic imagery PDF Children and deepfakes. At the collective scale, the harm is administrative and electoral: manipulation aimed at democratic processes PDF Future-Proofing Elections Against Deepfake Disinformation, and the broader machinery of amplified information manipulation now catalogued at scale PDF Information manipulation in the age of generative artificial intelligence. The citizen who “just shares” bears the reputational and legal risk; the platform that engineered the frictionless share does not.

What would help

The honest answer is modest. No literacy curriculum will let you eyeball a deepfake reliably — so the useful skill is not detection but provenance discipline: asking where a claim originated, who benefits, and whether it can be corroborated before it is forwarded PDF Déployer une littératie en IA pour une. Method beats model: building the habit of auditing a source’s chain of custody transfers across every new tool What building an auditable research tool taught me about …. The limitation is structural — no individual habit compensates for eroded platform safeguards. Literacy reduces your exposure; it does not repair the environment producing the risk.

Evidence Synthesis

Synthesizing this week’s 4,688 sources, the evidence on AI literacy points to an uncomfortable finding: the skill that matters most is no longer using AI but disbelieving it under adversarial conditions. This goes beyond technical competence. The literacy a citizen now needs is closer to a working theory of how manufactured doubt operates — who profits from your uncertainty, and when confident detection is itself a trap.

What the evidence shows

The convergent finding across serious analyses is that detection does not scale to the citizen. The Brennan Center’s assessment of whether AI fights or fuels election disinformation lands on a genuinely double-edged verdict — the same models that generate synthetic content also flag it, and neither side wins cleanly Does AI Fight or Fuel Election Disinformation?. The European Parliament’s brief on information manipulation reaches a parallel conclusion: generative tools lower the cost of falsehood faster than they lower the cost of verification PDF Information manipulation in the age of generative artificial intelligence. What actually builds resilience, per the Renaissance Numérique deployment report, is not a detection reflex but institutional literacy — knowing which sources carry accountability and which do not PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour une. CIVICUS’s election work echoes this: the intervention that held up was strengthening trusted intermediaries, not teaching individuals to spot pixel artifacts PDF Future-Proofing Elections Against Deepfake Disinformation.

Contested terrain

Here the evidence fractures. The most seductive form of literacy — “learn to detect fakes” — is where the failures cluster. Detection tools produce false positives at rates high enough to wrongly accuse real people, and a full 2026 guide documents how that error lands hardest on the innocent False Positives in AI Detection: Complete Guide 2026. Courts are already choking on this: judges report AI-generated evidence they cannot reliably adjudicate AI-generated evidence showing up in court alarms judges. So “literacy” splits in two — one camp treats it as a personal detection skill, the other as a structural question of which institutions get to certify truth. The Gartner-framed “battle for authority” names the real stake: not whether you can spot a deepfake, but who you’ll trust when you cannot Gartner has confirmed the battle for authority is happening right now. Is your agency ready for the aftermath?.

Across domains

Tool-specific literacy matters where the harm is intimate: voice cloning and identity fraud now demand that ordinary people know a verified phone call can be fabricated Deepfakes 2026: clonación de voz, fraude con IA y protección legal, and the Secret Service documents sextortion cases that weaponize exactly this credulity Sextortion in the Age of AI. As an equity issue, the burden is not evenly distributed — the OBVIA survey of amplified disinformation shows manipulation targets the least-resourced audiences first PDF Désinformation amplifiée par l’IA : incidents médiatisés, régulations …. Platform rules pretend to fill the gap; OpenAI’s political-campaigning restrictions are a vendor drawing its own boundary and calling it protection Political Campaigning Restrictions.

Gaps and uncertainty

What we do not know is large. There is no reliable longitudinal evidence that any literacy program durably changes behavior at scale, and this week’s sources carry zero mapped contradictions and zero documented failure statistics — an absence that should itself make you cautious. The Swiss debate over AI in democratic deliberation remains open, not settled Quand l’IA s’invite dans le débat démocratique suisse.

For citizens

Two takeaways carry weight. Individually: stop trying to detect, start verifying provenance — trace claims to accountable institutions rather than trusting your own eye PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour u. Collectively: the authority problem cannot be solved by educating you harder. It requires institutions worth trusting — which is a political demand, not a personal skill.

References

1. AI-generated evidence showing up in court alarms judges
2. Algorithmic Accountability Toolkit - Amnesty International
3. Deepfakes 2026: clonación de voz, fraude con IA y protección legal
4. Does AI Fight or Fuel Election Disinformation?
5. Election AI Deleted 5.5M Voters While World Governed Deepfakes Instead
6. False Positives in AI Detection: Complete Guide 2026
7. Gartner has confirmed the battle for authority is happening right now
8. Impossible d’envoyer un agent IA en prison
9. Misinformation may get worse in 2024 election as safeguards erode | AP News
10. Municipales 2026 : IA, deepfakes et désinformation, la démocratie …
11. [PDF Children and deepfakes - europarl.europa.eu](https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/775855/EPRS_BRI(2025)
12. PDF Désinformation amplifiée par l’IA : incidents médiatisés, régulations …
13. PDF Future-Proofing Elections Against Deepfake Disinformation
14. [PDF Information manipulation in the age of generative artificial intelligence](https://www.europarl.europa.eu/RegData/etudes/BRIE/2025/779259/EPRS_BRI(2025)
15. PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour u
16. PDF RAPPORT OCTOBRE 2025 Déployer une littératie en IA pour une
17. Platform-Facilitated Grooming and AI Chatbots
18. Political Campaigning Restrictions
19. Quand l’IA s’invite dans le débat démocratique suisse
20. Quand l’IA s’invite dans le débat démocratique suisse
21. Sextortion in the Age of AI
22. The Dangers of Unregulated AI in Policing - Brennan Center for Justice
23. The queryable war: When AI turns ‘could’ into ‘must’
24. What building an auditable research tool taught me about …