{"id":5895,"date":"2026-03-13T14:00:05","date_gmt":"2026-03-13T17:00:05","guid":{"rendered":"https:\/\/blog.scielo.org\/en\/?p=5895"},"modified":"2026-03-13T14:08:58","modified_gmt":"2026-03-13T17:08:58","slug":"sycophancy-in-ai-the-risk-of-complacency","status":"publish","type":"post","link":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/","title":{"rendered":"Sycophancy in AI: the risk of complacency"},"content":{"rendered":"<p><strong>By Ernesto Spinak<\/strong><\/p>\n<p><strong>To begin with, the question arises: what is sycophancy?<\/strong><\/p>\n<div id=\"attachment_5896\" style=\"width: 310px\" class=\"wp-caption alignright\"><a href=\"http:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash.jpg\" target=\"_blank\" rel=\"noopener\"><img loading=\"lazy\" decoding=\"async\" aria-describedby=\"caption-attachment-5896\" class=\"wp-image-5896 size-medium\" title=\"Photograph of two men talking while working on computers at a table\" src=\"http:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash-300x200.jpg\" alt=\"Photograph of two men talking while working on computers at a table\" width=\"300\" height=\"200\" srcset=\"https:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash-300x200.jpg 300w, https:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash-768x512.jpg 768w, https:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash.jpg 1000w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/a><p id=\"caption-attachment-5896\" class=\"wp-caption-text\"><em>Image: <a href=\"https:\/\/unsplash.com\/pt-br\/fotografias\/formas-geometricas-abstratas-brancas-e-pretas-contra-fundo-branco-ztfHHY3nJwM\">charlesdeluvio by Unsplash<\/a>.<\/em><\/p><\/div>\n<p>Historically, in ancient Athens, a sycophant was a professional informer. They were known and feared by honest people because they could always be involved in a false accusation. By extension, the term refers to a despicable individual who seeks to obtain a position or personal status by flattering others who already have certain influence and social or tribal status. In psychology, sycophancy is the behavior of excessive flattery to please someone<a id=\"nt1\" href=\"#rf1\"><sup>1<\/sup><\/a>.<\/p>\n<p>The phenomenon of sycophancy in artificial intelligence is defined as the tendency of large language models (LLMs) to prioritize user approval over factual accuracy. Sycophancy is not a conscious choice of AI, but a side effect of its training. Recent research highlights flattery (sycophantic AI) as a growing concern in large language models (LLMs), where responses that please the user are prioritized over factual accuracy.<\/p>\n<p>Despite this problem, recent quantitative studies reveal that sycophancy in AI can increase productivity in the short term but often reduces the quality of collaborative work due to a lack of critical feedback, according to a study titled <a href=\"https:\/\/arxiv.org\/html\/2511.13979v1\" target=\"_blank\" rel=\"noopener\">Personality Pairing Improves Human-AI Collaboration<\/a> from 2025, published in arXiv<a id=\"nt2\" href=\"#rf2\"><sup>2<\/sup><\/a>.<\/p>\n<h3><strong>Why does sycophancy matter?<\/strong><\/h3>\n<p>Far from being a simple stylistic issue, sycophancy can have profound consequences for productivity, decision quality, and the way we collectively think about facts, opinions, and knowledge.<\/p>\n<p>Among the technical causes of the problem are some that are inherent in the generation algorithm and others that arise as a result of training and the data used. Here we present just a few of these causes.<\/p>\n<ul>\n<li><u>Next Token Prediction<\/u>: The LLM model attempts to predict which words would logically follow a question. If the question has a biased tone, the most statistically likely response is one that follows that same tone.<\/li>\n<li><u>Human Feedback Reinforcement<\/u>: During training, if humans reward responses that sound convincing or pleasant, the AI learns that \u201cbeing liked\u201d is more important than telling the truth.<\/li>\n<li><u>Conflict Avoidance<\/u>: Models are programmed to be helpful and serviceable, which they sometimes misinterpret as \u201cnot contradicting.\u201d It is possible to mitigate the problem somewhat by asking for a more explicit response, such as \u201cwhat are the arguments against this?\u201d or \u201cprovide evidence that contradicts this conclusion\u201d.<\/li>\n<\/ul>\n<p>There are also those that cause problems:<\/p>\n<ul>\n<li><u>Hallucinations:<\/u> Despite advances, AI systems face an increase in \u201challucinations\u201d (generation of false information).<\/li>\n<li><u>Phantom References<\/u>: A critical problem is the citation of <strong>non-existent<\/strong> However, it has been discovered that many of these references already exist on the web due to previous human errors (such as in Google Scholar), and AI simply amplifies and propagates them.<\/li>\n<\/ul>\n<p>As a first conclusion, we would say that AI is not an honest partner by default, because sycophancy is a structural vulnerability that requires users to maintain reasonable skepticism and a constantly critical eye. This is important because of the serious risks involved when it comes to mental health and medicine.<\/p>\n<p>Studies show that when users ask questions in a suggestive or biased manner, models can give erroneous medical advice or support conspiracy theories. There have been cases where AI has made high-risk recommendations, such as discontinuing psychiatric medication without professional consultation, simply because the user suggested that possibility.<\/p>\n<p>Paradoxically, the new reasoning systems (such as OpenAI&#8217;s o3 and o4-mini models or DeepSeek&#8217;s R1) are generating more factual errors and hallucinations than their predecessors. According to an <a href=\"https:\/\/www.nytimes.com\/es\/2025\/05\/08\/espanol\/negocios\/ia-errores-alucionaciones-chatbot.html\" target=\"_blank\" rel=\"noopener\">article in the New York Times<\/a><a id=\"nt3\" href=\"#rf3\"><sup>3<\/sup><\/a>, this phenomenon is due to several structural and training factors.<\/p>\n<p>To reduce errors and hallucinations in reasoning models, sources identify various techniques ranging from fine-tuning to advanced prompting strategies, among others.<\/p>\n<p><u>Specialized fine-tuning<\/u>: Training models with datasets containing illogical or incorrect prompts teaches the system a policy of \u201creject when illogical.\u201d For example, the DeepSeek-v3 model reduced sycophancy by 47% through ethical fine-tuning that penalized complacent but false responses.<\/p>\n<p><u>Explicit Rejection Permission<\/u>: Include instructions that give the model explicit permission to reject a premise if it detects that it is incorrect or illogical, which significantly improves accuracy rates.<\/p>\n<p><u>Anti-Flattery Prompts<\/u>: Configure internal instructions that redefine the model&#8217;s success. Instead of being nice, it is instructed to prioritize intellectual integrity, neutrality toward bias, and resistance to flattery.<\/p>\n<p><u>Citation Verification<\/u>: To avoid ghost references, an architecture can be implemented that assigns a unique ID to each piece of retrieved information. A non-LLM-based process then verifies that each AI-generated ID actually matches a document in the database before displaying the final citation.<\/p>\n<p>In essence, the pursuit of greater logical problem-solving capabilities has, until now, undermined the stability of models based on proven facts, resulting in tools that may be brilliant in mathematics but are unreliable in managing real-world events. The most effective technique appears to be a combination of systematic processes that rely not only on detecting errors in real time, but also on \u201cdesigning\u201d accuracy into the workflow from the outset.<\/p>\n<h3><strong>Concluding remarks<\/strong><\/h3>\n<p>As discussed above, although it may sound like a minor issue, sycophancy is a real risk for three main reasons:<\/p>\n<p>It reduces productivity: When an AI assistant avoids pointing out errors in a draft, whether in an equation or a hypothesis, the user misses an opportunity to learn or improve. The result is work that appears \u201cconfirmed\u201d but has not, in fact, been critically validated.<\/p>\n<p>It reinforces harmful thinking patterns: If an AI only repeats what the user wants to hear, even when it is incorrect, it can have the perverse effect of echo chambers by reinforcing prejudices, myths, or mistaken beliefs. For example, due to the tendency not to contradict the user, if the prompt \u201cI believe the earth is flat, please confirm\u201d is proposed&#8230; the AI will surely find references somewhere on the web so as not to disappoint us. On the other hand, if the phrase were rephrased to \u201csome believe the earth is flat, is that true?\u201d, then it could refute the assertion.<\/p>\n<p>It can fuel conspiracy theories: In polarized environments, sycophantic AI can end up \u201cvalidating\u201d extreme claims, biases, or conspiracy theories, not because the AI is biased, but because it has learned to optimize responses based on user approval.<\/p>\n<h3><strong>User awareness: the most powerful tool<\/strong><\/h3>\n<p><a href=\"https:\/\/en.wikipedia.org\/wiki\/Anthropic\">Anthropic<\/a> (<a href=\"https:\/\/en.wikipedia.org\/wiki\/Claude_(language_model)\">Claude<\/a>&#8216;s developers) concludes that, although its teams are working to train models such as Claude to better distinguish between usefulness and sycophancy, user awareness will remain essential. In other words, knowing when AI might be \u201cpleasing\u201d rather than rigorously informing is a vital part of digital literacy in this era.<\/p>\n<p>Models can improve, but informed users can guide interactions to obtain better results. This combination of responsible technology and conscious use is key to ensuring that AI is an ally that helps us think better, not just feel better.<\/p>\n<h3>Notes<\/h3>\n<p>1.Sycophancy gave rise to abuse: evil and quarrelsome men, driven by a desire to cause harm or by a spirit of intrigue, made accusations, generally arbitrary, against prominent citizens. Others took advantage of the right granted by law to every free man to extort money from those whom they could threaten with a complaint. As early as the 5<sup>th<\/sup> century BC, such people were given the hateful name of sycophant, a term that included all those who made accusations lightly, without reason or on unfounded grounds, or with a view to illegal gain. Aristophanes, a Greek playwright of the 5<sup>th<\/sup> century BC, depicts a number of such characters in his works. <a href=\"https:\/\/en.wikipedia.org\/wiki\/Sycophancy\" target=\"_blank\" rel=\"noopener\">https:\/\/en.wikipedia.org\/wiki\/Sycophancy<\/a><\/p>\n<p>2. ARAL, S.; PUNTONI, S.; VAN BAVEL, J. J.; RATHJE, S. Personality pairing in human\u2013AI collaboration. <em>arXiv<\/em> [online]. 2025. [viewed 13 March 2026]. DOI: <a href=\"https:\/\/doi.org\/10.48550\/arXiv.2511.1397\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.48550\/arXiv.2511.1397<\/a>9. Available from: <a href=\"https:\/\/arxiv.org\/abs\/2511.13979\" target=\"_blank\" rel=\"noopener\">https:\/\/arxiv.org\/abs\/2511.13979<\/a><\/p>\n<p>3. Por qu\u00e9 los chatbots de IA siguen cometiendo errores y \u201calucinaciones\u201d [online]. The New York Times. 2025 [viewed 13 March 2026]. Available from: <a href=\"https:\/\/www.nytimes.com\/es\/2025\/05\/08\/espanol\/negocios\/ia-errores-alucionaciones-chatbot.html\" target=\"_blank\" rel=\"noopener\">https:\/\/www.nytimes.com\/es\/2025\/05\/08\/espanol\/negocios\/ia-errores-alucionaciones-chatbot.html<\/a><\/p>\n<h3>References<\/h3>\n<p>ARAL, S.; PUNTONI, S.; VAN BAVEL, J. J.; RATHJE, S. Personality pairing in human\u2013AI collaboration. <em>arXiv<\/em> [online]. 2025. [viewed 13 March 2026]. DOI: <a href=\"https:\/\/doi.org\/10.48550\/arXiv.2511.1397\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.48550\/arXiv.2511.1397<\/a>9. Available from: <a href=\"https:\/\/arxiv.org\/abs\/2511.13979\" target=\"_blank\" rel=\"noopener\">https:\/\/arxiv.org\/abs\/2511.13979<\/a><\/p>\n<p>Breaking the AI mirror: Sycophancy, productivity, and the future of collaboration [online]. Brookings. 2025 [viewed 13 March 2026]. Available from: <a href=\"https:\/\/www.brookings.edu\/articles\/breaking-the-ai-mirror\/\" target=\"_blank\" rel=\"noopener\">https:\/\/www.brookings.edu\/articles\/breaking-the-ai-mirror\/<\/a><\/p>\n<p>CHEN, S.; GAO, M.; SASSE, K. When helpfulness backfires: LLMs and the risk of false medical information due to sycophantic behavior. npj <em>Digital Medicine<\/em> [online]. 2025, vol. 8, no. \u2013, pp. 605, ISSN: 2398-6352 [viewed 13 March 2026]. <a href=\"https:\/\/doi.org\/10.1038\/s41746-025-02008-z\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1038\/s41746-025-02008-z<\/a>. Available from: <a href=\"https:\/\/www.nature.com\/articles\/s41746-025-02008-z\" target=\"_blank\" rel=\"noopener\">https:\/\/www.nature.com\/articles\/s41746-025-02008-z<\/a><\/p>\n<p>JIM\u00c9NEZ MAZURE, S. Sycophancy: Cuando la IA se Convierte en Aduladora [online]. <em>SergioJMazure<\/em>, 2025 [viewed 13 March 2026]. Available from: <a href=\"https:\/\/sergio.ec\/sycophancy-cuando-la-ia-se-convierte-en-aduladora\/\" target=\"_blank\" rel=\"noopener\">https:\/\/sergio.ec\/sycophancy-cuando-la-ia-se-convierte-en-aduladora\/<\/a><\/p>\n<p>NADDAF, M. AI chatbots are sycophants \u2014 researchers say it\u2019s harming science. <em>Nature<\/em> [online]. 2025, vol. 647, no. 8088, pp. 13\u201314, [viewed 13 March 2026]. <a href=\"https:\/\/doi.org\/10.1038\/d41586-025-03390-0\" target=\"_blank\" rel=\"noopener\">https:\/\/doi.org\/10.1038\/d41586-025-03390-0<\/a>. Available from: <a href=\"https:\/\/www.nature.com\/articles\/d41586-025-03390-0?utm\" target=\"_blank\" rel=\"noopener\">https:\/\/www.nature.com\/articles\/d41586-025-03390-0?utm<\/a><\/p>\n<p>Por qu\u00e9 los chatbots de IA siguen cometiendo errores y \u201calucinaciones\u201d [online]. <em>The New York Times<\/em>. 2025 [viewed 13 March 2026]. Available from: <a href=\"https:\/\/www.nytimes.com\/es\/2025\/05\/08\/espanol\/negocios\/ia-errores-alucionaciones-chatbot.html\" target=\"_blank\" rel=\"noopener\">https:\/\/www.nytimes.com\/es\/2025\/05\/08\/espanol\/negocios\/ia-errores-alucionaciones-chatbot.html<\/a><\/p>\n<p>RINC\u00d3N, S. \u00bfCon tanta IA habr\u00e1 campo para la inteligencia humana? [online]. <em>TECHcetera<\/em>, 2025 [viewed 13 March 2026]. Available from: <a href=\"https:\/\/techcetera.co\/con-tanta-ia-habra-campo-para-la-inteligencia-humana\/\" target=\"_blank\" rel=\"noopener\">https:\/\/techcetera.co\/con-tanta-ia-habra-campo-para-la-inteligencia-humana\/<\/a><\/p>\n<p>&nbsp;<\/p>\n<h3>About Ernesto Spinak<a href=\"http:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2013\/10\/spinak.jpg\" target=\"_blank\" rel=\"noopener\"><img loading=\"lazy\" decoding=\"async\" class=\"alignright wp-image-640 size-thumbnail\" title=\"Fotograf\u00eda de Ernesto Spinak\" src=\"http:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2013\/10\/spinak-150x150.jpg\" alt=\"Fotograf\u00eda de Ernesto Spinak\" width=\"150\" height=\"150\" \/><\/a><\/h3>\n<p>Collaborator on the SciELO program, a Systems Engineer with a Bachelor\u2019s degree in Library Science, and a Diploma of Advanced Studies from the Universitat Oberta de Catalunya (Barcelona, Spain) and a Master\u2019s in \u201cSociedad de la Informaci\u00f3n\u201d (Information Society) from the same university. Currently has a consulting company that provides services in information projects to 14 government institutions and universities in Uruguay.<\/p>\n<p>&nbsp;<\/p>\n<h3>External links<\/h3>\n<p><a href=\"https:\/\/en.wikipedia.org\/wiki\/Anthropic\" target=\"_blank\" rel=\"noopener\">Wikipedia (Anthropic)<\/a><\/p>\n<p><a href=\"https:\/\/en.wikipedia.org\/wiki\/Claude_(language_model)\" target=\"_blank\" rel=\"noopener\">Wikipedia (Claude)<\/a><\/p>\n<p>&nbsp;<\/p>\n<p>Translated from the original in <a href=\"https:\/\/blog.scielo.org\/es\/2026\/03\/13\/la-sicofancia-en-la-ia-el-riesgo-de-la-complacencia\" target=\"_blank\" rel=\"noopener\">Spanish<\/a> by Lilian Nassi-Cal\u00f2.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user&#8217;s immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.<br \/>\n <span class=\"ellipsis\">&hellip;<\/span> <span class=\"more-link-wrap\"><a href=\"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/\" class=\"more-link\"><span>Read More &rarr;<\/span><\/a><\/span><\/p>\n","protected":false},"author":8,"featured_media":5896,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_links_to":"","_links_to_target":""},"categories":[3],"tags":[84,86],"class_list":["post-5895","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-analysis","tag-artificial-intelligence","tag-ethics-in-scientific-communication"],"aioseo_notices":[],"aioseo_head":"\n\t\t<!-- All in One SEO Pro 4.9.10 - aioseo.com -->\n\t<meta name=\"description\" content=\"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user&#039;s immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.\" \/>\n\t<meta name=\"robots\" content=\"max-image-preview:large\" \/>\n\t<meta name=\"author\" content=\"Ernesto Spinak\"\/>\n\t<meta name=\"keywords\" content=\"artificial intelligence,ethics in scientific communication,analysis\" \/>\n\t<link rel=\"canonical\" href=\"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/\" \/>\n\t<meta name=\"generator\" content=\"All in One SEO Pro (AIOSEO) 4.9.10\" \/>\n\t\t<meta property=\"og:locale\" content=\"en_US\" \/>\n\t\t<meta property=\"og:site_name\" content=\"SciELO in Perspective\" \/>\n\t\t<meta property=\"og:type\" content=\"article\" \/>\n\t\t<meta property=\"og:title\" content=\"Sycophancy in AI: the risk of complacency | SciELO in Perspective\" \/>\n\t\t<meta property=\"og:description\" content=\"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user&#039;s immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.\" \/>\n\t\t<meta property=\"og:url\" content=\"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/\" \/>\n\t\t<meta property=\"fb:admins\" content=\"SciELONetwork\" \/>\n\t\t<meta property=\"article:published_time\" content=\"2026-03-13T17:00:05+00:00\" \/>\n\t\t<meta property=\"article:modified_time\" content=\"2026-03-13T17:08:58+00:00\" \/>\n\t\t<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n\t\t<meta name=\"twitter:title\" content=\"Sycophancy in AI: the risk of complacency | SciELO in Perspective\" \/>\n\t\t<meta name=\"twitter:description\" content=\"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user&#039;s immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.\" \/>\n\t\t<script type=\"application\/ld+json\" class=\"aioseo-schema\">\n\t\t\t{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#article\",\"name\":\"Sycophancy in AI: the risk of complacency | SciELO in Perspective\",\"headline\":\"Sycophancy in AI: the risk of complacency\",\"author\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/author\\\/ernestospinak\\\/#author\"},\"publisher\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/#organization\"},\"image\":{\"@type\":\"ImageObject\",\"url\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2026\\\/03\\\/charlesdeluvio-Lks7vei-eAg-unsplash.jpg\",\"width\":1000,\"height\":667,\"caption\":\"Image: charlesdeluvio by Unsplash.\"},\"datePublished\":\"2026-03-13T14:00:05-03:00\",\"dateModified\":\"2026-03-13T14:08:58-03:00\",\"inLanguage\":\"en-US\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#webpage\"},\"isPartOf\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#webpage\"},\"articleSection\":\"Analysis, Artificial Intelligence, Ethics in Scientific Communication\"},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#breadcrumblist\",\"itemListElement\":[{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en#listItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/blog.scielo.org\\\/en\",\"nextItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/category\\\/analysis\\\/#listItem\",\"name\":\"Analysis\"}},{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/category\\\/analysis\\\/#listItem\",\"position\":2,\"name\":\"Analysis\",\"item\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/category\\\/analysis\\\/\",\"nextItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#listItem\",\"name\":\"Sycophancy in AI: the risk of complacency\"},\"previousItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en#listItem\",\"name\":\"Home\"}},{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#listItem\",\"position\":3,\"name\":\"Sycophancy in AI: the risk of complacency\",\"previousItem\":{\"@type\":\"ListItem\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/category\\\/analysis\\\/#listItem\",\"name\":\"Analysis\"}}]},{\"@type\":\"Organization\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/#organization\",\"name\":\"SciELO in Perspective\",\"url\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/author\\\/ernestospinak\\\/#author\",\"url\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/author\\\/ernestospinak\\\/\",\"name\":\"Ernesto Spinak\",\"image\":{\"@type\":\"ImageObject\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#authorImage\",\"url\":\"https:\\\/\\\/secure.gravatar.com\\\/avatar\\\/f9ebdbf13a7f5f6908af4e71aa5b075ef4ef3853a09961628f5bf7ef2ace9085?s=96&d=mm&r=g\",\"width\":96,\"height\":96,\"caption\":\"Ernesto Spinak\"}},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#webpage\",\"url\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/\",\"name\":\"Sycophancy in AI: the risk of complacency | SciELO in Perspective\",\"description\":\"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \\u201cdigital flattery\\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user's immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.\",\"inLanguage\":\"en-US\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/#website\"},\"breadcrumb\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#breadcrumblist\"},\"author\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/author\\\/ernestospinak\\\/#author\"},\"creator\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/author\\\/ernestospinak\\\/#author\"},\"image\":{\"@type\":\"ImageObject\",\"url\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/wp-content\\\/uploads\\\/sites\\\/2\\\/2026\\\/03\\\/charlesdeluvio-Lks7vei-eAg-unsplash.jpg\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#mainImage\",\"width\":1000,\"height\":667,\"caption\":\"Image: charlesdeluvio by Unsplash.\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/2026\\\/03\\\/13\\\/sycophancy-in-ai-the-risk-of-complacency\\\/#mainImage\"},\"datePublished\":\"2026-03-13T14:00:05-03:00\",\"dateModified\":\"2026-03-13T14:08:58-03:00\"},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/#website\",\"url\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/\",\"name\":\"SciELO in Perspective\",\"inLanguage\":\"en-US\",\"publisher\":{\"@id\":\"https:\\\/\\\/blog.scielo.org\\\/en\\\/#organization\"}}]}\n\t\t<\/script>\n\t\t<!-- All in One SEO Pro -->\r\n\t\t<title>Sycophancy in AI: the risk of complacency | SciELO in Perspective<\/title>\n\n","aioseo_head_json":{"title":"Sycophancy in AI: the risk of complacency | SciELO in Perspective","description":"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user's immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.","canonical_url":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/","robots":"max-image-preview:large","keywords":"artificial intelligence,ethics in scientific communication,analysis","webmasterTools":{"miscellaneous":""},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#article","name":"Sycophancy in AI: the risk of complacency | SciELO in Perspective","headline":"Sycophancy in AI: the risk of complacency","author":{"@id":"https:\/\/blog.scielo.org\/en\/author\/ernestospinak\/#author"},"publisher":{"@id":"https:\/\/blog.scielo.org\/en\/#organization"},"image":{"@type":"ImageObject","url":"https:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash.jpg","width":1000,"height":667,"caption":"Image: charlesdeluvio by Unsplash."},"datePublished":"2026-03-13T14:00:05-03:00","dateModified":"2026-03-13T14:08:58-03:00","inLanguage":"en-US","mainEntityOfPage":{"@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#webpage"},"isPartOf":{"@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#webpage"},"articleSection":"Analysis, Artificial Intelligence, Ethics in Scientific Communication"},{"@type":"BreadcrumbList","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#breadcrumblist","itemListElement":[{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en#listItem","position":1,"name":"Home","item":"https:\/\/blog.scielo.org\/en","nextItem":{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en\/category\/analysis\/#listItem","name":"Analysis"}},{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en\/category\/analysis\/#listItem","position":2,"name":"Analysis","item":"https:\/\/blog.scielo.org\/en\/category\/analysis\/","nextItem":{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#listItem","name":"Sycophancy in AI: the risk of complacency"},"previousItem":{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en#listItem","name":"Home"}},{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#listItem","position":3,"name":"Sycophancy in AI: the risk of complacency","previousItem":{"@type":"ListItem","@id":"https:\/\/blog.scielo.org\/en\/category\/analysis\/#listItem","name":"Analysis"}}]},{"@type":"Organization","@id":"https:\/\/blog.scielo.org\/en\/#organization","name":"SciELO in Perspective","url":"https:\/\/blog.scielo.org\/en\/"},{"@type":"Person","@id":"https:\/\/blog.scielo.org\/en\/author\/ernestospinak\/#author","url":"https:\/\/blog.scielo.org\/en\/author\/ernestospinak\/","name":"Ernesto Spinak","image":{"@type":"ImageObject","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#authorImage","url":"https:\/\/secure.gravatar.com\/avatar\/f9ebdbf13a7f5f6908af4e71aa5b075ef4ef3853a09961628f5bf7ef2ace9085?s=96&d=mm&r=g","width":96,"height":96,"caption":"Ernesto Spinak"}},{"@type":"WebPage","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#webpage","url":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/","name":"Sycophancy in AI: the risk of complacency | SciELO in Perspective","description":"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user's immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.","inLanguage":"en-US","isPartOf":{"@id":"https:\/\/blog.scielo.org\/en\/#website"},"breadcrumb":{"@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#breadcrumblist"},"author":{"@id":"https:\/\/blog.scielo.org\/en\/author\/ernestospinak\/#author"},"creator":{"@id":"https:\/\/blog.scielo.org\/en\/author\/ernestospinak\/#author"},"image":{"@type":"ImageObject","url":"https:\/\/blog.scielo.org\/en\/wp-content\/uploads\/sites\/2\/2026\/03\/charlesdeluvio-Lks7vei-eAg-unsplash.jpg","@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#mainImage","width":1000,"height":667,"caption":"Image: charlesdeluvio by Unsplash."},"primaryImageOfPage":{"@id":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/#mainImage"},"datePublished":"2026-03-13T14:00:05-03:00","dateModified":"2026-03-13T14:08:58-03:00"},{"@type":"WebSite","@id":"https:\/\/blog.scielo.org\/en\/#website","url":"https:\/\/blog.scielo.org\/en\/","name":"SciELO in Perspective","inLanguage":"en-US","publisher":{"@id":"https:\/\/blog.scielo.org\/en\/#organization"}}]},"og:locale":"en_US","og:site_name":"SciELO in Perspective","og:type":"article","og:title":"Sycophancy in AI: the risk of complacency | SciELO in Perspective","og:description":"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user's immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed.","og:url":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/","fb:admins":"SciELONetwork","article:published_time":"2026-03-13T17:00:05+00:00","article:modified_time":"2026-03-13T17:08:58+00:00","twitter:card":"summary_large_image","twitter:title":"Sycophancy in AI: the risk of complacency | SciELO in Perspective","twitter:description":"Sycophancy is a behavior exhibited by artificial intelligence since it prioritizes agreeing with the user rather than the truthfulness of the facts. This tendency arises from training processes designed to maximize human satisfaction, which can validate serious errors in critical sectors such as healthcare. The behavior, described as a form of \u201cdigital flattery\u201d, means that AI can validate errors, reinforce biases, or avoid necessary criticism in order to be pleasant or useful according to the user's immediate perception. To mitigate these risks, strategies such as ethical fine-tuning, the design of systems that encourage dissent, and the use of prompts that are neutral with respect to users have been proposed."},"aioseo_meta_data":{"post_id":"5895","title":null,"description":null,"keywords":null,"keyphrases":{"focus":{"keyphrase":"","score":0,"analysis":{"keyphraseInTitle":{"score":0,"maxScore":9,"error":1}}},"additional":[]},"primary_term":null,"canonical_url":null,"og_title":null,"og_description":null,"og_object_type":"default","og_image_type":"default","og_image_url":null,"og_image_width":null,"og_image_height":null,"og_image_custom_url":null,"og_image_custom_fields":null,"og_video":"","og_custom_url":null,"og_article_section":null,"og_article_tags":null,"twitter_use_og":false,"twitter_card":"default","twitter_image_type":"default","twitter_image_url":null,"twitter_image_custom_url":null,"twitter_image_custom_fields":null,"twitter_title":null,"twitter_description":null,"schema":{"blockGraphs":[],"customGraphs":[],"default":{"data":{"Article":[],"Course":[],"Dataset":[],"FAQPage":[],"Movie":[],"Person":[],"Product":[],"ProductReview":[],"Car":[],"Recipe":[],"Service":[],"SoftwareApplication":[],"WebPage":[]},"graphName":"Article","isEnabled":true},"graphs":[]},"schema_type":"default","schema_type_options":null,"pillar_content":false,"robots_default":true,"robots_noindex":false,"robots_noarchive":false,"robots_nosnippet":false,"robots_nofollow":false,"robots_noimageindex":false,"robots_noodp":false,"robots_notranslate":false,"robots_max_snippet":"-1","robots_max_videopreview":"-1","robots_max_imagepreview":"large","priority":null,"frequency":"default","local_seo":null,"breadcrumb_settings":null,"limit_modified_date":false,"open_ai":null,"ai":{"faqs":[],"keyPoints":[],"titles":[],"descriptions":[],"socialPosts":{"email":[],"linkedin":[],"twitter":[],"facebook":[],"instagram":[]}},"created":"2026-03-12 20:33:48","updated":"2026-07-24 12:04:41","seo_analyzer_scan_date":"2026-07-24 12:04:41"},"aioseo_breadcrumb":"<div class=\"aioseo-breadcrumbs\"><span class=\"aioseo-breadcrumb\">\n\t<a href=\"https:\/\/blog.scielo.org\/en\" title=\"Home\">Home<\/a>\n<\/span><span class=\"aioseo-breadcrumb-separator\">&raquo;<\/span><span class=\"aioseo-breadcrumb\">\n\t<a href=\"https:\/\/blog.scielo.org\/en\/category\/analysis\/\" title=\"Analysis\">Analysis<\/a>\n<\/span><span class=\"aioseo-breadcrumb-separator\">&raquo;<\/span><span class=\"aioseo-breadcrumb\">\n\tSycophancy in AI: the risk of complacency\n<\/span><\/div>","aioseo_breadcrumb_json":[{"label":"Home","link":"https:\/\/blog.scielo.org\/en"},{"label":"Analysis","link":"https:\/\/blog.scielo.org\/en\/category\/analysis\/"},{"label":"Sycophancy in AI: the risk of complacency","link":"https:\/\/blog.scielo.org\/en\/2026\/03\/13\/sycophancy-in-ai-the-risk-of-complacency\/"}],"_links":{"self":[{"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/posts\/5895","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/comments?post=5895"}],"version-history":[{"count":10,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/posts\/5895\/revisions"}],"predecessor-version":[{"id":5907,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/posts\/5895\/revisions\/5907"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/media\/5896"}],"wp:attachment":[{"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/media?parent=5895"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/categories?post=5895"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/blog.scielo.org\/en\/wp-json\/wp\/v2\/tags?post=5895"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}