The best AI models cite retracted papers, and they cannot know it
Leading AI models frequently cite retracted academic papers without recognizing the retractions, as they lack real-time access to updated research databases.
- AI models cite retracted papers due to reliance on outdated training data.
- Recent retractions are missed entirely because models lack real-time updates.
- A live registry lookup system is proposed as the only viable solution.
- This issue raises concerns about AI's reliability in scientific and medical contexts.
A new analysis shows that frontier AI models often cite retracted academic papers without any awareness of the retractions. The study found that while models correctly flag older retractions they encountered during training, they miss recent ones entirely. This happens because AI systems rely on static snapshots of research databases, which do not reflect the latest corrections or retractions. The only reliable solution, according to the author, is a registry lookup system that cross-references citations against a live database of retracted papers. Without such a mechanism, AI models will continue to propagate outdated or discredited research, posing risks for accuracy in scientific and medical applications.
Source: The best AI models cite retracted papers, and they cannot know it. Read the full piece at the source.
AI systems need mechanisms to verify citations against live databases to avoid propagating retracted research.
Companies using AI for research or decision-making must implement safeguards to prevent reliance on outdated or discredited sources.
Students and researchers should be aware that AI-generated citations may include retracted papers, requiring manual verification.
AI's inability to recognize retracted research highlights a broader challenge in handling evolving knowledge.
- retracted papers
- Academic papers that have been formally withdrawn by their authors or publishers due to errors, misconduct, or other issues.
- training cutoff
- The date after which an AI model's training data no longer includes new information, limiting its knowledge to that point in time.
AI ResearchProxy-Pointer RAG: Temporal Reasoning Without Semantic Precompilation
AI ResearchClaude's hidden inner monologue is now readable thanks to Anthropic's new Jacobian Lens
Intelligent Computing appoints Philip S. Yu and Osmar R. Zaïane as new editors-in-chief - EurekAlert!
Addressing the AI Black Box Gap in Spectral Analysis - Spectroscopy Online
AI ResearchIdentifying Microbes in Space
HardwareAI Innovators Adopt NVIDIA Vera — Why Max Single-Threaded CPU at Scale Matters
NVIDIA introduces Vera, a new class of max single-threaded CPUs designed to accelerate agentic AI workflows by improving reasoning, response times, and tool execution.
Google, Anthropic, OpenAI lead AI Safety Index; SpaceXAI receives F - Seeking Alpha
Major AI labs Google, Anthropic, and OpenAI rank highly in an AI safety index, while SpaceXAI receives a failing grade.
UCSD aims to put AI research into practice - Healthcare IT News
The University of California San Diego is launching a new initiative to translate AI research into practical healthcare applications.
AI ToolsZero-Shot Local Document Parsing with Gemma 4: Treating PDFs as Images
Google’s Gemma 4 model can now parse PDFs by treating them as images, eliminating the need for text extraction pipelines.
SecurityI enabled Android's new security feature that detects fake cell towers - here's why
Android’s latest security feature can detect fake cell towers and suspicious networks, but users must opt in to receive alerts.
Ph.D. student finds AI can detect warning signs of global oil crises before they happen - Kennesaw State University
A Ph.D. student at Kennesaw State University developed an AI model that detects early warning signs of global oil crises before they occur.