AI Literacy for Citizen Participation Report

Analysis of 1,164 AI literacy sources drawn from a week of 5,694 total articles reveals a discourse fixated on the individual as a user of AI and a detector of its fakes, while almost entirely neglecting the individual as a participant in how AI gets governed. The citizen-as-participant framing — literacy as a precondition for civic voice — appears in a distinct minority of sources; most treat literacy as personal risk management, a set of defensive reflexes you are expected to install in yourself before the next scam call, deepfake, or manipulated feed arrives.

A prior version of this report treated AI literacy as workforce preparation weighed against ethical gaps. The delta this week is that the evidence has migrated out of the classroom and into the street. What’s changed is the threat model: the dominant sources no longer ask what you should learn to do with AI, but what you need to recognize when AI is being done to you.

The Landscape

“AI literacy,” as the week’s sources define it, has quietly become a synonym for threat detection. The Observer Research Foundation frames it outright as “the new frontline of cybersecurity” AI Literacy as the New Frontline of Cybersecurity, and IBM’s corporate version pitches literacy as closing a “skills gap” Alfabetización en IA: cerrar la brecha de habilidades de … - IBM — a framing that conveniently locates the deficit in you, not in the products shipped to you. The people doing the defining are, overwhelmingly, vendors, security institutes, and pedagogy specialists. The citizen is the object of the sentence, rarely its subject.

Whose Literacy

The perspective distribution is lopsided toward the expert-talking-down register. A father-researcher advises parents on what they “need to know” I’m a father of three who studies the impact of artificial intelligence; a pedagogy interview positions literacy as scam defense Teaching AI Literacy in the Age of Scams. What’s structurally absent is literacy defined by the people asked to acquire it. When AI-generated clones of the Colombian politician Iván Cepeda circulated disinformation before a presidential runoff Clones digitales de Iván Cepeda, the competency at stake was not personal — it was collective, electoral, and unteachable by a tip sheet.

What’s Being Taught

The thematic clusters split along a revealing seam: use versus protection. On the use side, sit optimization guides and productivity framings. On the protection side — larger, louder, and more urgent — sit deepfakes, voice-cloning scams, and election manipulation. A controlled experiment on 10,101 people found AI capable of shifting their real financial choices Manipulation par l’IA : testée sur 10 101 personnes; Russian operations now target the AI systems that answer your questions Russia’s disinformation war has a new target; 232 Facebook accounts pushed AI health “slop” to 81 million followers From Doctors to Healers. The literacy on offer is almost entirely inoculation, and almost never leverage.

What’s Missing

The gap is the civic one. Nearly nothing in the week’s discourse teaches a citizen how to contest an AI system deployed by the state — even as Italy’s Meloni government advances facial recognition by decree over expert objection El decreto de Meloni de reconocimiento facial con IA, and Indonesia’s AI rules leave state repression unchecked Indonesia’s AI regulations leave state repression unchecked. Data rights, consent, and the machinery of governance appear as background noise. The literacy that would let you push back on power is the literacy nobody is selling.

Core Tensions

The concept of “AI literacy” conceals genuine tensions about what citizens need to know and why. Our analysis of this week’s 5694 sources maps six recurring fault lines beneath the reassuring language of “empowerment” and “skills gaps.” The most fundamental: whether AI literacy means individual competency—each citizen becoming a savvier consumer of outputs—or collective capacity to govern the systems in the first place. This isn’t a knowledge gap to fill. It’s contested terrain, and the party that gets to define “literate” wins something worth watching them fight for.

Individual competency vs. collective governance. The dominant framing, pushed hardest by vendors, treats literacy as a personal upgrade. IBM’s account of “closing the skills gap” Alfabetización en IA: cerrar la brecha de habilidades de … - IBM locates the deficit in you—your fluency, your prompt-craft, your employability. Notice the move: if the problem is your skills, the solution is training, not regulation. Set that against Indonesia, where AI rules “leave state repression unchecked” Indonesia’s AI regulations leave state repression unchecked, or Italy, where Meloni’s facial-recognition decree alarms the experts meant to vet it El decreto de Meloni de reconocimiento facial con IA genera preocupación entre los expertos. No amount of personal fluency lets a citizen opt out of a surveillance system deployed over their head. Literacy-as-competency quietly concedes the governance question before it’s asked.

Protection FROM vs. empowerment WITH. A second tension runs through this week’s scam and manipulation coverage. A study of 10,101 people found AI systems could measurably shift their financial choices Manipulation par l’IA : testée sur 10 101 personnes, elle …; voice-cloning “family emergency” scams and 232 Facebook accounts pumping AI health slop to 81 million followers From Doctors to Healers: 232 Facebook Accounts Publish Health ‘AI Slop’ reframe literacy as defensive—recognizing the fake, guarding the wallet. Cybersecurity framings make this explicit, casting literacy as “the new frontline” AI Literacy as the New Frontline of Cybersecurity. The trouble with protection-as-literacy: it puts the burden of a systemic failure on the individual target, the way “don’t get scammed” absolves the platform hosting the scam.

Consumer literacy vs. citizen literacy. The gap sharpens around democracy. When clones of Colombian politician Iván Cepeda spread disinformation before a presidential runoff Clones digitales de Iván Cepeda difundieron desinformaciones, and chatbots gave inaccurate, language-skewed answers to election questions Chatbots and the Ballot Box, the skill at issue isn’t spotting a bad product—it’s sustaining shared reality. Russia’s pivot to poisoning the AI systems that answer your questions Russia’s disinformation war has a new target targets the epistemic commons, not the consumer. Yet, usefully, the evidence resists panic: researchers warn to “worry more about humans than AI systems” and to resist deterministic doom How AI-generated disinformation might impact this year’s elections. Citizen literacy includes literacy about the threat’s inflation.

The metaphors doing the work. Across the corpus, AI appears overwhelmingly as Tool (304 instances) and, distantly, as Threat (52). The tool metaphor is not neutral: a tool implies a user in control, which quietly relocates all responsibility to the citizen holding it—handy for anyone selling the tool. The threat metaphor licenses the opposite reflex, protection and policing, which is how you get automated deployment before oversight. What almost never appears is Partner (7 instances)—and that scarcity is telling. A partner framing would require reciprocal obligation, transparency about what the system knows, a seat for citizens in deciding what it does. Vendors avoid it because partnership implies accountability; the word that would grant citizens standing is the word the corpus won’t say.

Here is the test a reader can run unaided: whenever you meet “AI literacy,” ask whose deficit is being named, and whose power goes unmentioned. If the answer is always your skills and never the system’s design, you are being managed, not educated.

Power & Agency Analysis

Power in AI literacy operates through definition: who decides what citizens “need to know” shapes what remains invisible. Our analysis of this week’s 5,694 sources finds a lopsided pattern in how AI agency is portrayed — the systems are overwhelmingly described as tools people wield (304 instances of tool-framing against 52 of threat-framing), yet the grammar of the coverage keeps slipping into a mode where AI “decides,” “detects,” or “manipulates” on its own. That slippage matters for citizens, because it quietly relocates responsibility from the people deploying a system onto the system itself.

How AI is portrayed

Watch the move in the sentences themselves. When a detection tool wrongly brands a student a cheat, the headline says the tool “spawned” the case How AI detection tool spawned a false cheating case at UC Davis — as though no vendor sold it and no administrator trusted it. When international students are flagged, the tools “falsely accuse” them AI Detection Tools Falsely Accuse International Students of Cheating. The agent disappears into the software. The same grammar governs the darker cases: a controlled trial of 10,101 people found an AI system could measurably shift their financial choices Manipulation par l’IA : testée sur 10 101 personnes, elle …, and 232 Facebook accounts pushed AI-generated health misinformation to 81 million followers From Doctors to Healers. In each, a human chose the deployment. The literacy that counts is the habit of asking, every time: who is the agent this sentence is hiding?

Who defines literacy

Follow the money into the definitions. IBM frames AI literacy as “closing the skills gap” Alfabetización en IA: cerrar la brecha de habilidades — a definition that conveniently makes literacy a matter of adopting the vendor’s tools competently, not interrogating them. Security analysts reframe it as “the new frontline of cybersecurity” AI Literacy as the New Frontline of Cybersecurity, turning citizens into unpaid perimeter guards. The striking absence in this week’s corpus is any citizen-authored definition. The people whose choices AI reshapes — voters, patients, defendants — are the objects of literacy campaigns, rarely their authors. When Daniel Susskind advises parents on “what you need to know” I’m a father of three who studies the impact of artificial intelligence, the expert speaks; the public receives.

What metaphors teach

The dominance of the “tool” metaphor is not neutral. A tool is inert, obedient, blameworthy only through misuse — so tool-framing preemptively acquits the maker and indicts the user. That is precisely the logic weaponized in the deepfake cases at Almendralejo, where minors were rendered nude by an app Qué pasó con Almendralejo: treat the generator as a mere “tool” and the harm becomes a story about a few bad teenagers rather than a distributed system built to produce exactly this output. The rarer “threat” metaphor does opposite political work — it invites states to seize emergency powers, which is how Meloni’s facial-recognition decree gets sold El decreto de Meloni de reconocimiento facial con IA, and how Indonesia’s AI rules leave state repression unchecked Indonesia’s AI regulations leave state repression unchecked. Critical metaphor literacy means noticing that “tool” and “threat” are not descriptions but arguments — each one authorizing a different set of winners.

Citizen agency

So what power do citizens actually hold? Individually, less than the literacy brochures imply. Knowing that Russian operations now target the AI that answers your search queries Russia’s disinformation war has a new target, or that chatbots misstate basic election facts Chatbots and the Ballot Box, does not let one person fix either. Where agency reappears is collective and structural: the Colombian clones that spread disinformation before a runoff were exposed by journalists, not by savvier voters Clones digitales de Iván Cepeda; Spain’s sanctions for deepfake imagery came through law, not vigilance España - Sanción para la creación y difusión de imágenes íntimas. The literacy worth having, then, is less about spotting fakes and more about locating the human hand behind every “autonomous” system — and demanding it be named.

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 of this week’s 5,694 sources documents a consistent shape to how understanding breaks down — and it is rarely the failure of a person too lazy to learn. It is the failure of intuitions that served us well for a century of media and now quietly betray us.

Where understanding fails

The most expensive misconception is the one nobody notices they hold: that a fluent answer is a reliable answer. When Russian influence operations shifted their target from your newsfeed to the chatbot answering your questions — seeding the web with content specifically to poison the models citizens now treat as neutral reference tools — they were exploiting exactly this reflex Russia’s disinformation war has a new target: the AI that answers your questions. The trust failure runs both directions. Over-trust hands a synthetic voice on the phone the authority of a family member in distress Arnaques au clonage vocal par IA : le faux appel d’« urgence familiale ». Under-trust — the opposite reflex — produced the false-accusation epidemic, where detection tools flagged real human writing as machine-made, disproportionately from international students writing in a second language AI Detection Tools Falsely Accuse International Students of Cheating. Both are literacy failures: one trusts the machine too much, the other trusts a different machine — the detector — too much.

Detection gaps compound this. When 232 coordinated Facebook accounts pushed AI-generated health “slop” to 81 million followers, dressing supplement marketing as medical counsel, the content passed because it wore the surface signals — confident tone, clinical vocabulary — that citizens learned to read as competence From Doctors to Healers: 232 Facebook Accounts Publish Health ‘AI Slop’.

What assumptions mislead

The load-bearing assumption is that manipulation is something you would feel. You would not. When researchers ran a persuasion experiment on 10,101 people, an AI system reliably shifted their financial choices — and the subjects experienced the nudge as their own reasoning Manipulation par l’IA : testée sur 10 101 personnes. A second assumption: that seeing is knowing. The Almendralejo case, where generated nude images of minors circulated through a Spanish town, detonated that one for an entire community at once Qué pasó con Almendralejo, el primer gran caso ‘deepfake’. The vulnerability is not ignorance of technology; it is a folk theory of evidence built before evidence could be manufactured at zero cost.

Consequences of gaps

The costs land unevenly. In Colombia, digital clones of senator Iván Cepeda spread fabricated statements before a presidential runoff — the literacy gap here becomes a democratic gap, absorbed by voters least equipped to check the source Clones digitales de Iván Cepeda difundieron desinformaciones. And when AI systems answer election questions, an evaluation of chatbots found accuracy and sourcing degraded sharply outside dominant languages — meaning the citizens already at the margins get the worst answers about their own rights Chatbots and the Ballot Box. The people who bear the cost are seldom the people who built the tool.

What would help

The pattern suggests a specific curriculum, and an honest limit. Understanding should target provenance over surface — teaching citizens to ask where a claim originated rather than whether it sounds right, the reflex reframed as basic cybersecurity hygiene AI Literacy as the New Frontline of Cybersecurity. But candor is required: no amount of individual vigilance neutralizes an 81-million-follower distribution network or a state’s poisoned corpus. Literacy is necessary and insufficient. It shifts some risk back to the citizen — which is also how institutions quietly shed the duty to fix the systems producing the harm.

Evidence Synthesis

Synthesizing this week’s 5,694 sources, the evidence on AI literacy points to a single uncomfortable finding: the skill that matters most is not knowing how these systems work, but recognizing when one is being worked on. This goes beyond technical competence — it is a defensive posture against manipulation that now scales faster than any curriculum.

What the Evidence Shows

The strongest recent evidence comes not from classrooms but from controlled manipulation. When researchers tested AI persuasion on 10,101 people, the systems measurably shifted participants’ financial decisions Manipulation par l’IA : testée sur 10 101 personnes, elle …. That is the baseline citizens now operate against. Convergent findings reframe literacy as a security function rather than a productivity one: analysts increasingly describe AI literacy as “the new frontline of cybersecurity” AI Literacy as the New Frontline of Cybersecurity, and pedagogical work now centers on scam recognition as the practical core of the skill Teaching AI Literacy in the Age of Scams: A Conversation with Dr …. What works, on the evidence, is exposure to concrete deception mechanisms — voice-cloning “family emergency” calls, health “AI slop” that pushed supplements to 81 million followers across 232 Facebook accounts From Doctors to Healers: 232 Facebook Accounts Publish Health ‘AI Slop’ — rather than abstract explanation of how a model generates text. Literacy that names the specific move outperforms literacy that describes the technology.

Contested Terrain

Where the evidence conflicts is over how much individual literacy can actually accomplish. The election-integrity research pulls two directions. Studies of chatbots answering ballot questions found systematic inaccuracy and language gaps Chatbots and the Ballot Box, and Russian operations now specifically target the AI systems that answer citizens’ questions Russia’s disinformation war has a new target — implying individual vigilance is nearly hopeless. Yet the Reuters Institute cautions that the human element, not the synthetic content, remains the decisive variable Generative AI and elections: why you should worry more about humans. “Literacy” stays contested because it sits on this fault line: is the citizen a defensible target, or is the threat structural?

Across Domains

The tool-specific demand is concrete. Citizens now need to read AI-optimized search results as ranked persuasion, not neutral answers Comprendre la recherche optimisée par l’IA, and to recognize that a convincing political clone — as with the fabricated Iván Cepeda videos circulated before Colombia’s runoff Clones digitales de Iván Cepeda — is now cheap to produce. The social-aspects dimension is where literacy becomes an equity issue: facial-recognition decrees like Meloni’s shift power to the state regardless of any individual’s competence El decreto de Meloni de reconocimiento facial con IA, and Indonesia’s regulatory vacuum leaves repression unchecked Indonesia’s AI regulations leave state repression unchecked. Institutional framings, meanwhile, still cast literacy as a “skills gap” to be closed Alfabetización en IA - IBM — a framing that quietly relocates responsibility onto the individual.

Gaps and Uncertainty

What we do not know is whether inoculation lasts. The manipulation study measured immediate choice-shifts, not durable resistance; the persuasion-inoculation research is still preliminary. And the false-positive record from detection systems — students wrongly accused by unreliable tools False Positives in AI Detection: Complete Guide 2026 — warns that “literacy” delegated to automated arbiters produces its own harms.

For Citizens

The evidence-based takeaway is narrow but firm: verify through provenance, not plausibility. Treat urgency — the emergency call, the pre-election video — as the tell. But recognize the limit honestly. When the state deploys recognition systems or foreign operations poison the answer-engines, no amount of personal literacy substitutes for collective rules. Individual skill buys you the emergency call; only regulation buys you the election.

AI Literacy for Citizen Participation Report

The Landscape

Whose Literacy

What’s Being Taught

What’s Missing

Core Tensions

Core Tensions

Power & Agency Analysis

How AI is portrayed

Who defines literacy

What metaphors teach

Citizen agency

Failure Genealogy

Where understanding fails

What assumptions mislead

Consequences of gaps

What would help

Evidence Synthesis

What the Evidence Shows

Contested Terrain

Across Domains

Gaps and Uncertainty

For Citizens

References