Grover is a sophisticated transformer-based model developed by the Allen Institute for AI (AI2) specifically designed to both generate and detect neural fake news. Unlike general-purpose LLMs, Grover is architected to understand the nuances of journalistic style, including metadata fields such as domain, date, authors, and headlines. This multidimensional approach allows it to achieve state-of-the-art accuracy in identifying machine-written text. Its core technical premise is that the best defense against neural fake news is a model that can generate it; thus, Grover's detector is simply the Grover generator itself fine-tuned for classification. As of 2026, Grover remains a foundational pillar in the AI forensics space, widely used by social media platforms, fact-checking organizations, and security researchers to identify synthetic propaganda. Its architecture enables it to detect outputs from diverse generators including GPT-4, Llama-4, and Claude iterations, making it a critical tool in the fight against industrial-scale disinformation. The model is released in three sizes: Base, Large, and Mega (1.5 billion parameters), allowing for a balance between inference speed and detection precision.
Analyzes the consistency between text body and metadata fields like 'Domain' and 'Author' to find anomalies.
Verified feedback from the global deployment network.
Post queries, share implementation strategies, and help other users.
A 1.5B parameter model that outperforms smaller classifiers in identifying sophisticated LLM outputs.
The detector is built by training on the generator's own output, creating a self-improving feedback loop.
Trained to detect broad statistical patterns found in transformer architectures, not just Grover's own output.
Extracts stylistic markers specific to automated news generators.
Supports top-p (nucleus) sampling for generation, allowing researchers to simulate various levels of AI sophistication.
Outputs detailed confidence scores for every part of an article, from headline to body.
Identifying bot networks spreading machine-generated fake news articles in real-time.
Registry Updated:2/7/2026
Editors receiving automated press releases or 'citizen' reports that look too polished to be human.
Quantifying the risk of future LLMs being used to flood the internet with untraceable fake news.