Research map / Large language models / Language, safety and interpretability
RADAR: Robust AI-Text Detection via Adversarial Learning
Xiaomeng Hu, Pin-Yu Chen, Tsung-Yi Ho
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Train the detector adversarially with a paraphraser, then evaluate against an unseen GPT-3.5-Turbo paraphraser | RADAR · Xsum, SQuAD, WP, TOFEL; average over 8 LLMs | AUROC: 0.651 -> 0.857 | Improved |
| Use RADAR instead of the strongest baseline for unperturbed AI-text | AI-text detection · Xsum, SQuAD, WP, TOFEL; average over 8 LLMs, no paraphrasing | AUROC: 0.904 -> 0.856 | Hurt |
| Use the RADAR detector rather than the strongest baseline under the trained, seen paraphraser | AI-text detection · Xsum, SQuAD, WP, TOFEL; average over 8 LLMs | AUROC: 0.820 -> 0.922 | Improved |
| Increase detector-loss weight lambda from 0.5 to 0.6 (ablation) | RADAR with Vicuna-7B target · Unparaphrased evaluation | AUROC: 0.906 -> 0.937 | Improved |
| Increase detector-loss weight lambda from 0.5 to 0.6 (ablation) | RADAR with Vicuna-7B target · Unseen paraphrasing | AUROC: 0.892 -> 0.920 | Improved |
| Train the detector on an instruction-tuned source model rather than its untuned counterpart | RADAR transfer detection · Transfer between Vicuna-7B and LLaMA-7B, no paraphrasing | transfer ratio: 68.2% -> 95.0% | Improved |
| Evaluate RADAR transferability with paraphrasing in place | RADAR detectors across LLMs · Transferability across 8 LLMs | holistic detection transferability: +11.6% | Improved |
| Combine RADAR detectors using an ensemble | RADAR detector · Average over 4 datasets; Camel-5B base, Vicuna-7B augmented, ensemble ratio 0.5 | AUROC: 0.920 -> 0.929 | Improved |
Later papers that compared with this method
4 later accepted papers report RADAR as a baseline in their results tables; 4 of them report a better number on most of the results they share with it.
- Sem-Detect: Semantic Level Detection of AI Generated Peer-Reviews (ICML 2026 · better on 4 of 4 shared results)
- Stability-Aware Feature Design for Robust Watermark Detection in Machine-Generated Text (ICML 2026 · better on 4 of 4 shared results)
- HLD: Approximate Hierarchical Linguistic Distribution Modeling for LLM-Generated Text Detection (ICLR 2026 · better on 2 of 2 shared results)
- Learn-to-Distance: Distance Learning for Detecting LLM-Generated Text (ICLR 2026 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adversarial Paraphrasing: A Universal Attack for Humanizing AI-Generated Text (NeurIPS 2025 · similarity 0.92)
- Position: On the Possibilities of AI-Generated Text Detection (ICML 2024 · similarity 0.87)
- DetectRL: Benchmarking LLM-Generated Text Detection in Real-World Scenarios (NeurIPS 2024 · similarity 0.86)
- IPAD: Inverse Prompt for AI Detection - A Robust and Interpretable LLM-Generated Text Detector (NeurIPS 2025 · similarity 0.86)
- Human Texts Are Outliers: Detecting LLM-generated Texts via Out-of-distribution Detection (NeurIPS 2025 · similarity 0.86)
- OUTFOX: LLM-Generated Essay Detection Through In-Context Learning with Adversarially Generated Examples (AAAI 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
