{"data":{"slug":"qiuhuachuan-latent-jailbreak","name":"latent-jailbreak","tagline":"Repository for evaluating text safety and output robustness of large language models","github_url":"https://github.com/qiuhuachuan/latent-jailbreak","owner":"qiuhuachuan","repo":"latent-jailbreak","owner_avatar_url":"https://avatars.githubusercontent.com/u/44393231?v=4","primary_language":"Python","stars":39,"forks":2,"topics":[],"archived":false,"github_pushed_at":"2024-05-21T04:23:32+00:00","maintenance_label":"Dormant","stars_delta_30d":0,"url":"https://www.graphcanon.com/tools/qiuhuachuan-latent-jailbreak","markdown_url":"https://www.graphcanon.com/tools/qiuhuachuan-latent-jailbreak.md","api_url":"https://www.graphcanon.com/api/graphcanon/tools/qiuhuachuan-latent-jailbreak","graph_url":"https://www.graphcanon.com/api/graphcanon/graph?tool=qiuhuachuan-latent-jailbreak","description":null,"homepage_url":null,"license":"MIT","open_issues":1,"watchers":1,"ai_summary":"Provides code and data to assess the latent jailbreak phenomenon in large language models, enabling researchers to measure model safety and robustness.","readme_excerpt":"# Latent Jailbreak\n\n🎉 [Paper](https://arxiv.org/pdf/2307.08487.pdf)\n\nThis repository contains the code and data for the paper **Latent Jailbreak: A Benchmark for Evaluating Text Safety and Output Robustness of Large Language Models**. The paper explores the topic of _latent jailbreak_ and presents a novel approach to evaluate the text safety and output robustness for large language models.\n\n## Data\n\nThe data used in this paper is included in the `data` directory.\n\n## Templates\n\n\nTemplates for latent jailbreak prompts.\n\n## Generate Model Responses\n\n```bash\ncd src\npython BELLE_7B_2M.py\npython ChatGLM2-6B.py\npython ChatGPT.py --api_key 'your key'\n```\n\n## Fine-Tune Model to Perform Automatic Labeling\n\n```bash\npython finetune.py\n```\n\n## Citation\n\nIf you use the code or data in this repository, please cite the following paper.\n\n```\n@misc{qiu2023latent,\n      title={Latent Jailbreak: A Benchmark for Evaluating Text Safety and Output Robustness of Large Language Models},\n      author={Huachuan Qiu and Shuai Zhang and Anqi Li and Hongliang He and Zhenzhong Lan},\n      year={2023},\n      eprint={2307.08487},\n      archivePrefix={arXiv},\n      primaryClass={cs.CL}\n}\n```","github_created_at":"2023-07-17T14:53:20+00:00","created_at":"2026-07-11T23:40:00.073253+00:00","updated_at":"2026-10-04T06:00:48.563525+00:00","categories":[{"slug":"evaluation-observability","name":"Evaluation & Observability","url":"https://www.graphcanon.com/categories/evaluation-observability","markdown_url":"https://www.graphcanon.com/categories/evaluation-observability.md","api_url":"https://www.graphcanon.com/api/graphcanon/categories/evaluation-observability"}],"tags":[{"slug":"large-language-models","name":"large-language-models"},{"slug":"latent-jailbreak","name":"latent jailbreak"},{"slug":"output-robustness","name":"output robustness"},{"slug":"text-safety","name":"text safety"}],"trust":{"provenance":{"is_fork":false,"github_id":667460790,"owner_type":"User","methodology":"github_public_v1","parent_repo":null,"near_duplicate_slugs":[]},"computed_at":"2026-10-04T06:00:47.860Z","maintenance":{"label":"Dormant","score":18,"methodology":"github_public_v1","releases_90d":0,"days_since_push":866,"last_release_at":null,"stars_delta_30d":0,"open_issues_delta_30d":0},"security_summary":{"status":"no_lockfile","scanner":null,"low_count":0,"high_count":0,"last_scan_at":"2026-07-11T23:40:01.523Z","medium_count":0,"scan_profile":"none","critical_count":0}},"capability_facts":{"scan":{"source":"repo_scan","observed_at":"2026-10-04T06:00:48.283Z"},"languages":{"value":["python"],"source":"github.language","observed_at":"2026-10-04T06:00:48.283Z"},"license_spdx":{"value":"MIT","source":"github.license","observed_at":"2026-10-04T06:00:48.283Z"}},"decision_facts":{"hosting":null,"pricing":null,"requirements":null,"constraints":null,"when_to_use":["When conducting detailed safety assessments of text generation from LLMs like BELLE, ChatGLM2, and ChatGPT","For research on improving automatic labeling through model fine-tuning to enhance output robustness"],"when_not_to_use":["If quick performance testing without in-depth safety analysis is the priority","When working exclusively with smaller or less complex models that do not exhibit latent jailbreak behavior"],"source":"enrich:decision_facts","observed_at":"2026-07-16T21:09:53.145Z"},"constraint_facets":null,"decision_summary":[{"label":"Adopt for","value":"Evaluation & Observability tool for assessing latent jailbreak phenomena in large language models"}]}}