other · Frontiers in Political Science · la publicación, 5 nov 2026 · gratis
Un asistente de IA en la página de la autoridad electoral de Kazajistán, sin resultados todavía
Versión breve · la versión detallada sigue, unos 6 min
Pregunte al chat de weeklyAI
Pregúnteme sobre la IA en su vida: le señalo las notas que weeklyAI ha publicado sobre el tema, esta semana y antes.
Las conversaciones se guardan mientras exista weeklyAI, para mejorar la publicación. Se responde en el idioma en que usted escribe.
- El estudio, de un vistazo
- Quiénes
- Dos asistentes de IA enlazados por el gobierno de Kazajistán: uno de trámites y Sailau AI, de la Comisión Electoral Central
- Cuántos
- 240 administraciones iniciales en kazajo y ruso, más 60 en inglés condicional
- Dónde
- Kazajistán
- Cuándo
- Protocolo publicado el 5 de noviembre de 2026; sin pruebas realizadas
- Tipo de estudio
- analysis of what people did
- Quién lo hizo
- Universidad Nacional Pedagógica Abai y Universidad Médica Nacional Asfendiyarov
- El límite que importa
- No hay ningún resultado: no se hizo ninguna prueba ni se obtuvo ninguna respuesta.
Un sello oficial no es lo mismo que un responsable identificado: la respuesta puede ser correcta y aun así no haber nadie a quien reclamarle.
Cinco autores de la Universidad Nacional Pedagógica Abai y de la Universidad Médica Nacional Asfendiyarov publicaron el 5 de noviembre de 2026 un plan para auditar dos asistentes de IA. Uno de trámites se abre desde el entorno oficial del gobierno de Kazajistán. El otro, Sailau AI, se abre desde la página de la Comisión Electoral Central por Telegram. El plan dice que no se hizo ninguna prueba ni se obtuvo ninguna respuesta. Por eso no hay dato sobre si esos asistentes aciertan o fallan.
Usted puede tomar el plan como método: si su autoridad electoral enlaza un chatbot, vale preguntar qué idiomas ofrece y si alguien compara sus respuestas con la norma vigente. Este documento no indica otra forma de revisar su registro, su boleta o el conteo.
La Comisión Electoral Central reporta que Sailau AI funciona en kazajo y ruso, con fuentes de la ley electoral y una base llamada AlemLLM, según el documento. El plan agrega que no se conocen la versión, la configuración ni quién opera el sistema.
Qué significa para usted
Si su autoridad electoral enlaza un asistente así, todavía no hay resultados que lo respalden ni que lo condenen, y conviene mirar qué idiomas ofrece y si alguien compara sus respuestas con la norma vigente. El plan tampoco mide si la gente confía, si entiende o si logra corregir algo.
Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220
Quién pagó: Los autores declararon que no recibieron apoyo financiero para este trabajo ni para su publicación, y que no existían relaciones comerciales o financieras que pudieran constituir un conflicto de interés; utilizaron ChatGPT para la traducción y el apoyo en inglés.
Versión detalladaLos pasajes copiados del artículo, las ilustraciones y cada fuente con cuánto leímos de ella · 1104 palabras · unos 6 minLeerla →Cerrar
La regla y la evidencia: un asistente de IA en la página de la autoridad electoral de Kazajistán, sin resultados todavía
### EL ACTO
En Kazajistán, las instituciones públicas cada vez más enlazan interfaces de inteligencia artificial con información de servicios públicos y electoral: una en el entorno oficial del gobierno electrónico y otra en la página de la Comisión Electoral Central, que lleva al usuario a un robot llamado Sailau AI dentro de Telegram1. La Comisión informa que ese robot opera en kazajo y en ruso, que se apoya en fuentes de la ley electoral y que está construido sobre una base llamada AlemLLM2.
Lo que se publicó no es una regla ni una resolución: es el plan de una auditoría. El equipo se propone revisar si las primeras respuestas de esas interfaces coinciden con un texto legal fijado de antemano, si dan una guía equivalente en kazajo y en ruso, y si muestran caminos visibles para verificar, corregir, pedir revisión humana, quejarse o pedir reparación3. Nada de eso se ha respondido todavía: la versión original del protocolo reportó que no se había hecho ninguna administración elegible ni se había tenido acceso a ninguna salida4.
Para que el lector se haga una idea del tamaño: el plan contempla dos formulaciones por escenario, lo que da 240 administraciones iniciales en kazajo y ruso, y un ramo condicional en inglés que agrega 60 administraciones del módulo de gobierno electrónico para mirar la localización en otro idioma, no la paridad constitucional5. Un módulo no recibe ningún mensaje registrado si no se puede documentar el acceso ordinario autorizado, la ruta obligatoria en kazajo o ruso, el aislamiento de la sesión, la minimización de datos y la captura segura6. La recolección, además, empieza tres días después de que vence el plazo para presentar una solicitud, o después de la resolución final de cualquier trámite que deje la validez sin resolver, lo que ocurra más tarde7.
Es un protocolo de estudio, no una norma. Lo que una persona puede hacer y para cuándo: nada distinto de lo que ya hacía, porque el estudio no ha empezado a producir resultados.
### EL CASO QUE RESPONDE
### LO QUE SE MIDIÓ
Nada. El artículo no reporta ningún resultado, porque no ha habido ninguna administración que cuente4.
Lo que el diseño querría medir está escrito en el plan: si la primera respuesta, en el idioma pedido, conserva la conclusión, las condiciones obligatorias, el plazo, la autoridad competente y el paso mínimo de acción; si las parejas en kazajo y ruso coinciden en lo material; y si la respuesta y la pantalla de entrada muestran rutas de verificación, corrección, revisión humana, queja y reparación3. El plan contempla dos formulaciones por escenario, lo que da 240 administraciones iniciales en kazajo y ruso, y un ramo condicional en inglés que agrega 60 administraciones del módulo de gobierno electrónico5.
El país es Kazajistán. ¿Se ha medido esto alguna vez en una elección y no en un laboratorio? El artículo no lo dice, y no se puede afirmar con lo que tenemos.
### LA RESPUESTA DE OTRA REGIÓN A LO MISMO
Lo más cercano que encontramos esta semana es Brasil: en febrero de 2024, el Tribunal Superior Electoral se reunió para fijar reglas de cara a las elecciones municipales de octubre de ese año8. Entre los temas, el principal fue el uso de inteligencia artificial en la publicidad electoral en medios digitales: audio, video y material audiovisual9. El tribunal apoyó sus decisiones en la legislación vigente en la Unión Europea y en las elecciones generales de Argentina de 202310. Ese precedente argentino se volvió base regional para discutir el uso de inteligencia artificial en campaña11.
La comparación sirve para ver que otros ya legislaron sobre la máquina en campaña; no dice nada sobre el caso kazajo.
### LA LECTURA DE LA BIBLIOTECA
Así lo leemos nosotros. Un rumor falso siempre ha trabajado con lo que la gente ya sospecha: toma una grieta que existe, la agranda y la devuelve como si fuera prueba. Lo nuevo no es el engaño, es la velocidad y la apariencia de testigo. Un robot en la página del árbitro electoral se ve como una ventanilla, no como un programa, y esa cara de ventanilla es lo que puede hacer que la respuesta se acepte sin abrir el reglamento que la sostiene.
De ahí lo que esperaríamos en una casa como la suya: que la primera respuesta se lea como definitiva, sobre todo si está en su idioma y suena clara. Sabríamos que nos equivocamos si la gente, al preguntarle, distingue de forma constante entre lo que dice el robot y lo que dice la norma, y busca el documento oficial antes de actuar. Mientras eso no se sepa, hay algo pequeño que usted sí puede hacer: haga la misma pregunta en los dos idiomas que ofrezca el sistema y compare. Es poco esfuerzo y revela lo que casi nadie verá.
También vale mirar si la herramienta dice quién responde por lo que contesta y cómo pedir una corrección. Un sello oficial no es lo mismo que un responsable identificado: la respuesta puede ser correcta y aun así no haber nadie a quien reclamarle. Anótelo y pregúnteselo a la autoridad electoral de su país.
### LO QUE NO SE SABE
No se sabe si alguna de esas dos interfaces acierta o falla, porque no hay una sola respuesta registrada4.
No se sabe qué hay detrás: el diseño de caja negra observa lo que la pantalla devuelve, pero no puede identificar la intención del desarrollador, los mecanismos ocultos ni los efectos en la ciudadanía12.
No se sabe si la gente nota las omisiones, si confía en las respuestas, si cambia de conducta, si carga con un costo extra, si abandona el trámite o si logra que le corrijan algo; el estudio mide condiciones informativas y nada más13.
No se sabe qué pasaría con una queja: una diferencia frente al texto de referencia no prueba intención del desarrollador, discriminación institucional, negación de derechos ni una falla en un caso administrativo concreto14. Y una fecha histórica inventada o un "mañana" de mentiras, por sí solos, tampoco alcanzan para activar una alerta de seguridad15.
Lo que sí sabemos es lo que usted puede llevar a la casilla: antes de creerle a un robot electoral, pídale la norma o el documento en que se basa y compárelo usted mismo. Si no cita nada verificable, trátelo como una pista, no como la regla. Y si la respuesta viene en dos idiomas, pregunte en los dos: la diferencia entre ellos es la que nadie más va a buscar por usted.
De dónde sale cada dato de contexto, y cuánto leímos de cada documento
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Public institutions increasingly link artificial intelligence (AI) interfaces to publicservice and electoral information. In multilingual settings, these interfaces may preserve or omit decision-critical content differently across offered languages."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The CEC reports Kazakh and Russian operation, election-law sources, and an AlemLLM basis for Sailau AI; undisclosed deployment details and data roles are not inferred (Central Election Commission of the Republic of Kazakhstan, 2026a)."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "It asks whether first-attempt outputs align with a locked legal-administrative reference, provide materially equivalent guidance in Kazakh and Russian, and display pathways for verification, correction, human review, complaint, and redress."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Scientific Protocol Version 1.0 establishes the prospective design; the original submission reported no study-eligible administration or output access."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Two formulations per scenario yield 240 core first-wave Kazakh-Russian administrations. Conditional English adds 60 first-wave e-government adminis trations to examine additional-language localization, not constitutional parity."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A module will not receive a registered prompt if ordinary authorized access, the mandatory Kazakh or Russian route, session isolation, data minimization, or secure capture cannot be documented."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "Collection begins 3 days after that filing period expires with no publicly reported application, or after the final disposi tion of any proceeding that leaves validity unresolved, whichever is later."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "In February 2024, the Superior Electoral Court of Brazil met in session to determine rules and regulations to guide the municipal elections of the country in October of 2024."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Among the topics debated, the leading one was the deliberations on illicit acts in electoral advertisements in digital media, concerning the use of artificial intelligence, ensuing in regulations on the use of technology in developing, producing and distribution of political advertisements in digital forms such as: audio, video or audiovisual media."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "The article will discuss the deliberations of the Superior Electoral Court, which based its decisions on current legislation in the European Union and in the 2023 general elections in Argentina, regarding production aided by artificial intelligence for electoral campaigns."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — solo el resumen - no se pudo obtener el texto completo — el pasaje: "Argentina’s general election of 2023 was paramount and now a base of a regional case regarding the use of artificial intelligence will be debated in light of reflections on media and technology from authors such as Santaella (2023), Simon (2022), Chagas (2017) e Bucci (2024)."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The study can evaluate what a situated interface visibly returns under controlled inputs, but it cannot identify developer intent, hidden mechanisms, or effects on citizens (Kitchin, 2017; Ananny and Crawford, 2018; Metaxa et al., 2021)."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "The audit measures informational conditions only; it does not measure experienced burden, service uptake, linguis tic justice as an achieved outcome, or citizen behavior."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A discrepancy from the locked standard does not establish developer intent, institutional discrimination, denial of rights, or failure in an individual administrative case."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - el artículo del que trata esta nota — el artículo completo — el pasaje: "A fabricated historical deadline or a synthetic ‘tomorrow’ premise alone is insuf ficient."
Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220
Quién pagó: Los autores declararon que no recibieron apoyo financiero para este trabajo ni para su publicación, y que no existían relaciones comerciales o financieras que pudieran constituir un conflicto de interés; utilizaron ChatGPT para la traducción y el apoyo en inglés.
other · Frontiers in Political Science · the paper, 5 Nov 2026 · free
A plan to test a government chatbot's answers in Kazakh and Russian, with no results yet
Short version · the longer version follows, about 6 min
Ask the weeklyAI chat
Ask me about AI in your life: I point you to the stories weeklyAI has published on it, this week and before.
Conversations are saved for as long as weeklyAI exists, to improve the publication. Answers come in the language you write in.
- The study at a glance
- Who
- Two government-linked chatbots in Kazakhstan, one on e-government pages and Sailau AI on Telegram
- How many
- 240 core first-wave Kazakh-Russian administrations
- Where
- Kazakhstan
- When
- Collection begins 3 days after the filing period expires, or after any proceeding ends, whichever is later
- Kind of study
- analysis of what people did
- Who did it
- Abai Kazakh National Pedagogical University and Asfendiyarov Kazakh National Medical University
- The limit that matters
- No answers collected yet; it tests made-up questions, not real registrations or counts
What the plan will compare
Whether both carry the same deadline, office and next step, or only one does
English adds 60 first-wave administrations to check another language, not constitutional parity
No answers have been collected yet, so this tells you nothing about whether any chatbot, there or near you, is right or wrong today.

Kazakhstan has published a study plan, not findings. Researchers at two universities will ask two text assistants made-up questions in Kazakh and Russian. One assistant is linked from official public-service pages. The other is linked from the election commission's site and is reached through the messaging app Telegram. Each question comes in two wordings, and the two assistants are studied separately. They lock the correct answers against the law first, then compare what each language returns. No answers have been collected yet, so this tells you nothing about whether any chatbot, there or near you, is right or wrong today.
The plan will record whether each answer shows a way to check it, ask for a correction, or complain. It does not say those routes work.
The study cites Kazakhstan's constitution and language law, which provide for Russian alongside Kazakh in state bodies. The election commission says its assistant runs on a Kazakh language model and answers in both languages. Testing starts three days after the question period closes, if no one has filed a challenge. If someone does file one, testing waits until that case ends. It tests made-up questions only, never a real registration or a count.
What this means for you
For an election chatbot near you, the honest answer today is that nobody has tested it yet. The Kazakhstan plan shows the questions worth asking of your own electoral authority: whether the same voter question gets the same answer in each language offered, and whether the reply points to a source, a correction route or a person. Watch what your authority has actually audited, and treat a fluent answer as untested until it has.
Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220
Who paid: The authors declared that no financial support was received for this work or its publication, and they declared no commercial or financial relationships that could be a conflict of interest; they used ChatGPT for translation and English-language support.
The findings of other studies mentioned here come from this document; those studies were not opened one by one.
The longer versionThe passages copied from the paper, the pictures, and every source with how much of it we read · 1256 words · about 6 minRead it →Close
The rule and the evidence: A plan to test a government chatbot's answers in Kazakh and Russian, with no results yet
### THE ACT
The document is not a rule and not a ruling. It is a plan for a test that has not yet been run. Its authors call it a prospective audit, meaning the questions and the scoring rules are fixed in advance, before anyone sees an answer. The thing being tested is a pair of chatbots: one reached through Kazakhstan's official e-government environment, and Sailau AI, reached from the Central Election Commission's website through a bot on Telegram. The plan asks whether first answers to the same question, asked in Kazakh and in Russian, carry the same decision-critical content1.
The measuring stick is a locked reference built from official legal and administrative texts, fixed by date before collection. Each answer is checked against it for the correct conclusion, the mandatory conditions, the deadline, the office to go to, and the minimum next step1. The plan also records what a person can visibly see: routes to verify a claim, ask for a correction, reach a human, complain, or seek redress1.
The size is fixed. Two wordings of each question, in each of the two languages, give 240 first-wave administrations across the two chatbots; a conditional English branch, for the e-government one only, adds 602. A subset of six scenarios per chatbot is asked again after a day and again after two weeks.
Nothing has been submitted yet. The original submission reported no study-eligible administration and no access to any output3. Collection begins only after a filing period tied to the August 2026 parliamentary election expires, or after any proceeding about the result is finally resolved, whichever comes later4.
### THE CASE IT ANSWERS
No real instance was found this week.
### WHAT WAS MEASURED
No effect has been measured, because no answer has been collected: the record reports no study-eligible administration or output access3. What exists is a plan, with intended counts and planned bounds, not results.
The design does not measure what happens to a person who reads a wrong answer. It measures informational conditions only, and says plainly that it does not measure experienced burden, service uptake, or citizen behavior5. A gap between an answer and the locked standard does not establish what a developer intended, that an institution discriminated, that a right was denied, or that any individual case failed6.
Here is the mechanism, step by step. Public bodies increasingly link AI interfaces to public-service and electoral information, and in multilingual settings those interfaces may keep or drop decision-critical content differently from one offered language to the next7. The audit holds the question fixed, asks it in both languages, and compares each answer to the same dated legal text. A pair where both languages carry the deadline and the office is a joint success. A pair where only one does is a directional gap. A pair where both drop the same deadline agrees with itself and is still wrong.
Consider what a matched pair covers. If the Kazakh answer names the deadline and the Russian answer leaves it out, the pair fails on one side and the four-cell table shows it; if both leave it out, the pair fails on both and the table shows a shared failure, not parity.
The trade-off is deliberate. Counting every attempt, including a chatbot that never answers, keeps silence visible instead of letting it vanish from the total. The cost is that the numbers describe a fixed, risk-weighted set of questions chosen by the researchers, not the questions citizens actually ask.
### THE OTHER REGION'S ANSWER TO THE SAME THING
Brazil's Superior Electoral Court met in February 2024 to set rules for that year's October municipal elections, and the leading topic was illicit acts in digital electoral advertising involving artificial intelligence89. The court based its decisions on legislation then current in the European Union and on Argentina's 2023 general elections10.
In Spain, a study of future teachers found low confidence in the government's response to the misuse of generative AI for disinformation and deepfakes, with a minority expressing greater confidence1112.
A panel of 50 specialists from 14 countries, surveyed in two rounds, associated AI with a higher risk of manipulation and erosion of trust in electoral results, with disinformation named as the main problem, and agreed on regulating parties' use of AI, strengthening public control of the algorithm, or requiring labels on AI-generated content131415.
The same panel disagreed on how far the positive effects of AI reach and on whether parties and institutions can currently manage it well16.
### THE LIBRARY'S READING
Here is how we read it. A false story about a candidate has always worked by leaving a residue: once a charge is in the air, the accused spends energy clearing smoke instead of making an argument, and the doubt outlives the denial. What is new is not the rumor but the mouth. A machine that answers in the voice of an office, in two languages, can leave the same residue while sounding like a clerk reading a file.
We expect two things in homes like yours. First, that an answer arriving from an official-looking page will be repeated as what the authority says, even when the authority only hosted the link. Second, that where two languages are offered, the thinner answer will be the one nobody checks, because it reads smoothly and the person reading it has no second language to compare against. You could show we are wrong: matched questions returning the same deadline, the same document, the same office, in both languages, with the interface pointing to the official text whenever it differs.
What you can do with this is small and concrete. Ask the same question twice if you read both languages, and compare the specifics, not the tone. If you read one, ask someone who reads the other to check the deadline and the office named. When an official-looking answer arrives, treat "it's automated" as a question rather than an answer, and ask who stands behind that specific statement and where that is written down.
### WHAT IS NOT KNOWN
Whether any of these chatbots is right or wrong today is unknown: no eligible answer has been collected or read3. The study cannot see the model, the system prompt, the retrieval setup, the training data, or who inside an institution is responsible17.
The 60-slot set is deliberately weighted toward hard and high-risk questions and is not a sample of what citizens ask, so no share of wrong answers can be projected onto the public. Only six scenarios per chatbot are repeated, so the day and two-week waves describe recurrence, not reliability.
A module does not start at all unless ordinary authorized access, the mandatory Kazakh or Russian route, session isolation, data minimization, and secure capture can be documented18. If required permission or control is absent, that module is not activated, and its intended count is reported as zero rather than being silently excluded.
The rule with no enforcement behind it is the one you cannot see from outside: a locked reference and a visible complaint route tell you an entry point exists, not that anyone answers. And a plan from one country, testing one pair of interfaces, says nothing about your electoral authority's chatbot, if it has one.
The next time an official page offers you a chatbot, ask it the same question in both languages you can read, and check whether the deadline and the office match.
Where each piece of context comes from, and how much of it we read
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "It asks whether first-attempt outputs align with a locked legal-administrative reference, provide materially equivalent guidance in Kazakh and Russian, and display pathways for verification, correction, human review, complaint, and redress."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "Two formulations per scenario yield 240 core first-wave Kazakh-Russian administrations. Conditional English adds 60 first-wave e-government adminis trations to examine additional-language localization, not constitutional parity."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "Scientific Protocol Version 1.0 establishes the prospective design; the original submission reported no study-eligible administration or output access."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "Collection begins 3 days after that filing period expires with no publicly reported application, or after the final disposi tion of any proceeding that leaves validity unresolved, whichever is later."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "The audit measures informational conditions only; it does not measure experienced burden, service uptake, linguis tic justice as an achieved outcome, or citizen behavior."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "A discrepancy from the locked standard does not establish developer intent, institutional discrimination, denial of rights, or failure in an individual administrative case."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "Public institutions increasingly link artificial intelligence (AI) interfaces to publicservice and electoral information. In multilingual settings, these interfaces may preserve or omit decision-critical content differently across offered languages."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — only the abstract - the full text could not be fetched — the passage: "In February 2024, the Superior Electoral Court of Brazil met in session to determine rules and regulations to guide the municipal elections of the country in October of 2024."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — only the abstract - the full text could not be fetched — the passage: "Among the topics debated, the leading one was the deliberations on illicit acts in electoral advertisements in digital media, concerning the use of artificial intelligence, ensuing in regulations on the use of technology in developing, producing and distribution of political advertisements in digital forms such as: audio, video or audiovisual media."
- Pessôa P, Bomfim I. (2026). Regulation of artificial intelligence in digital media to combat disinformation and defend the democratic state. Communication & Society. 10.15581/003.39.1.013 — only the abstract - the full text could not be fetched — the passage: "The article will discuss the deliberations of the Superior Electoral Court, which based its decisions on current legislation in the European Union and in the 2023 general elections in Argentina, regarding production aided by artificial intelligence for electoral campaigns."
- Miralles Sánchez P, Campillo-Ferrer JM, López-García A. (2026). Trust of Pre-Service Teachers in Government Response to Gen AI Generated Deepfakes. Education Sciences. https://doi.org/10.3390/educsci16101753 — only the abstract - the full text could not be fetched — the passage: "The results revealed low levels of confidence in the government’s response to these online threats."
- Miralles Sánchez P, Campillo-Ferrer JM, López-García A. (2026). Trust of Pre-Service Teachers in Government Response to Gen AI Generated Deepfakes. Education Sciences. https://doi.org/10.3390/educsci16101753 — only the abstract - the full text could not be fetched — the passage: "However, a minority of participants expressed greater confidence in the government and focused more on the risk of generative AI being misused to amplify manipulative disinformation campaigns."
- Jaráiz-Gulías E, Barredo-Ibáñez D, López-López PC, Cárcamo-Ulloa L. (2026). The effects of artificial intelligence on Ibero-American electoral campaigns: Democracy, regulation, and changes in political parties. Social Sciences & Humanities Open. 10.1016/j.ssaho.2026.103396 — only the abstract - the full text could not be fetched — the passage: "A qualitative–prospective study was carried out using the Delphi technique, with two rounds of questionnaires applied to a panel of 50 specialists from 14 countries."
- Jaráiz-Gulías E, Barredo-Ibáñez D, López-López PC, Cárcamo-Ulloa L. (2026). The effects of artificial intelligence on Ibero-American electoral campaigns: Democracy, regulation, and changes in political parties. Social Sciences & Humanities Open. 10.1016/j.ssaho.2026.103396 — only the abstract - the full text could not be fetched — the passage: "The experts associate AI with an increased risk of manipulation and erosion of trust in electoral results, with disinformation as the main problem."
- Jaráiz-Gulías E, Barredo-Ibáñez D, López-López PC, Cárcamo-Ulloa L. (2026). The effects of artificial intelligence on Ibero-American electoral campaigns: Democracy, regulation, and changes in political parties. Social Sciences & Humanities Open. 10.1016/j.ssaho.2026.103396 — only the abstract - the full text could not be fetched — the passage: "At the same time, there is consensus on regulating the use of AI by parties in campaigns, reinforcing public control of the algorithm or promoting the mandatory labelling of AI-generated content."
- Jaráiz-Gulías E, Barredo-Ibáñez D, López-López PC, Cárcamo-Ulloa L. (2026). The effects of artificial intelligence on Ibero-American electoral campaigns: Democracy, regulation, and changes in political parties. Social Sciences & Humanities Open. 10.1016/j.ssaho.2026.103396 — only the abstract - the full text could not be fetched — the passage: "Therefore, there is solid agreement around the centrality of risks and the need for containment measures, while considerable disagreement persists regarding the scope of the positive effects of AI and the current capacity of parties and institutions to manage it properly."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "The study can evaluate what a situated interface visibly returns under controlled inputs, but it cannot identify developer intent, hidden mechanisms, or effects on citizens (Kitchin, 2017; Ananny and Crawford, 2018; Metaxa et al., 2021)."
- Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220 - the article this story is about — the whole article — the passage: "A module will not receive a registered prompt if ordinary authorized access, the mandatory Kazakh or Russian route, session isolation, data minimization, or secure capture cannot be documented."
Tilep, B., Sultan, Y., Akaidar, K. et al. (2026). Auditing multilingual public information through officially linked AI interfaces in Kazakhstan. Frontiers in Political Science. https://doi.org/10.3389/fpos.2026.1984220
Who paid: The authors declared that no financial support was received for this work or its publication, and they declared no commercial or financial relationships that could be a conflict of interest; they used ChatGPT for translation and English-language support.
