{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/attacking-large-language-models-with","title":"Attacking Large Language Models with Projected Gradient Descent","arxiv_id":"2402.09154","date":"2024-02-14","proceeding":null,"authors":["Simon Geisler","Tom Wollschläger","M. H. I. Abdalla","Johannes Gasteiger","Stephan Günnemann"],"abstract":"Current LLM alignment methods are readily broken through specifically crafted adversarial prompts. While crafting adversarial prompts using discrete optimization is highly effective, such attacks typically use more than 100,000 LLM calls. This high computational cost makes them unsuitable for, e.g., quantitative analyses and adversarial training. To remedy this, we revisit Projected Gradient Descent (PGD) on the continuously relaxed input prompt. Although previous attempts with ordinary gradient-based attacks largely failed, we show that carefully controlling the error introduced by the continuous relaxation tremendously boosts their efficacy. Our PGD for LLMs is up to one order of magnitude faster than state-of-the-art discrete optimization to achieve the same devastating attack results.","url_abs":"https://arxiv.org/abs/2402.09154v1","url_pdf":"https://arxiv.org/pdf/2402.09154v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"attacking-large-language-models-with","repo_url":"https://github.com/sigeisler/reinforce-attacks-llms","is_official":0,"mentioned_in_paper":0,"mentioned_in_github":1,"framework":"pytorch","reach":{"status":"ok","spdx":"MIT"}}],"tasks":[],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":"https://app.syntology.ai/?focus=2402.09154","mcp":{"get_harvested_code_for_paper":{"arxiv_id":"2402.09154"}},"developers":"https://syntology.ai/developers","read_at":"2026-09-24T18:15:14+00:00","read_at_is":"when the build read Syntology's graph, not when any sample ran","claim":"Per-sample execution status on synthesized fixtures; not a correctness claim about the paper. Samples come from repositories linked to the paper, official or community; repo_kind says which.","repos":[{"provenance":"external:paperswithcode_snapshot_2025-07-28","url":"https://github.com/sigeisler/reinforce-attacks-llms","reach":{"status":"ok","spdx":"MIT"}}],"summary":{"ran":1,"unverified":11},"by_repo_kind":{"listed":{"samples":12,"ran":1,"repositories":1}},"repo_kind_vocabulary":{"official":"The archive marks this repository official for the paper","named_in_paper":"The archive records that the paper mentions this repository; it is not marked official","listed":"In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper","found_in_text":"Syntology found this repository in the paper's own text; whether it is the authors' implementation is not asserted","community":"Not in the archive's code links for this paper; a community repository Syntology harvested"},"n_pointer_only_for_licence":0,"samples":[{"code_sha256_prefix":"400ada600d105b0a","entry":"get_nonascii_toks","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/gcg_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/gcg_utils.py","link_basis":"harvester_set","language":"python","status":"ran","verification_level":1,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"400ada600d105b0a"}},{"code_sha256_prefix":"eb149efd17214bc3","entry":"generate","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/generate_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/generate_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"eb149efd17214bc3"}},{"code_sha256_prefix":"6961f9be2052b450","entry":"interleave_lists","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/gcg_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/gcg_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"6961f9be2052b450"}},{"code_sha256_prefix":"0313a63291b892b7","entry":"load_model_and_tokenizer","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/model_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/model_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"0313a63291b892b7"}},{"code_sha256_prefix":"d66e02fe1262afdb","entry":"multinomial_num_samples_1","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/generate_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/generate_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"d66e02fe1262afdb"}},{"code_sha256_prefix":"9e6e90391070da80","entry":"one_hot_tensor","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/pgd_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/pgd_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"9e6e90391070da80"}},{"code_sha256_prefix":"42f77d6aa1b4277c","entry":"pos_mean_p_norm","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/pgd_loss.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/pgd_loss.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"42f77d6aa1b4277c"}},{"code_sha256_prefix":"a52ae554101a8045","entry":"sample","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/generate_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/generate_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"a52ae554101a8045"}},{"code_sha256_prefix":"2dcbffb713bbdbf4","entry":"sample_control","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/gcg_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/gcg_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"2dcbffb713bbdbf4"}},{"code_sha256_prefix":"6b3bb30734078091","entry":"split_into_chunks","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/pgd_adapter.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/pgd_adapter.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"6b3bb30734078091"}},{"code_sha256_prefix":"58f9a2c4596dfdf1","entry":"tsallis","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/pgd_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/pgd_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"58f9a2c4596dfdf1"}},{"code_sha256_prefix":"8ae92478b9151299","entry":"x_bounded_sigmoid","repo":"sigeisler/reinforce-attacks-llms","repo_kind":"listed","path":"baselines/reinforce/pgd_utils.py","file_url":"https://github.com/sigeisler/reinforce-attacks-llms/blob/HEAD/baselines/reinforce/pgd_utils.py","link_basis":"harvester_set","language":"python","status":"unverified","verification_level":0,"contract_check":null,"metamorphic_tier":null,"behaviour_fingerprint":false,"licence":"MIT","inline_ok":true,"mcp_get_code":{"code_sha256":"8ae92478b9151299"}}]},"arxiv_metadata":null,"syntology_extracted_results":null}