Papers › Revisiting Character-level Adversarial Attacks for Language Models

Revisiting Character-level Adversarial Attacks for Language Models

7 May 2024arXiv:2405.04346archive 2025-07-28

Elias Abad Rocamora, Yongtao Wu, Fanghui Liu, Grigorios G. Chrysos, Volkan Cevher

Adversarial attacks in Natural Language Processing apply perturbations in the character or token levels. Token-level attacks, gaining prominence for their use of gradient-based methods, are susceptible to altering sentence semantics, leading to invalid adversarial examples. While character-level attacks easily maintain semantics, they have received less attention as they cannot easily adopt popular gradient-based methods, and are thought to be easy to defend. Challenging these beliefs, we introduce Charmer, an efficient query-based adversarial attack capable of achieving high attack success rate (ASR) while generating highly similar adversarial examples. Our method successfully targets both small (BERT) and large (Llama 2) models. Specifically, on BERT with SST-2, Charmer improves the ASR in 4.84% points and the USE similarity in 8% points with respect to the previous art. Our implementation is available in https://github.com/LIONS-EPFL/Charmer.

PaperPDFCodeCode Syntology ranMethod as code

In Syntology View this paper on Syntology, its page in Syntology's graph. That page adds which of its functions are also present in other papers' code Syntology holds, with dates, and its dated run record as JSON and a README badge. It also lists the repositories linked to the paper, the abstract and the calls for agents.

Open this paper in Syntology's Atlas, the map of the papers in Syntology's graph and their citations.

For agents, on Syntology's MCP service (how to connect):

Code

Syntology 31 code samples harvested from 1 repository linked to this paper: 23 ran (of which 3 constructed an object rather than computing a result; 10 with no instrument failure: 2 honoured, 1 violated, 7 with no contract checked; 13 where Syntology's instrument failed) · 8 unverified. Where Syntology's instrument failed: 11 where our contract draft was wrong; 2 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: official repository: 31 samples from 1 repository, 23 ran (10 with no instrument failure; 13 where Syntology's instrument failed). The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

Levels written between 2026-09-22 and 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

lions-epfl/charmer officialmentioned in paperpytorchMITSyntology: 31 samples, 23 ran (13 instrument-failed) · 8 unverified report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

31 samples harvested: 23 ran (of which 3 constructed an object rather than computing a result; 10 with no instrument failure: 2 honoured, 1 violated, 7 with no contract checked; 13 where Syntology's instrument failed) · 8 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page last changed ; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

2ran · honoured contract
1ran · violated contract
11ran · our draft was wrong
2ran; the contract check could not run
7ran
8unverified

Licence: 0 of the 31 samples are pointer only, meaning Syntology does not serve that copy's text. This page shows no code text for any sample; each one links to its file in the repository.

Harvested from lions-epfl/charmer. “Ran” means the sample executed on a synthesized input. It does not mean the output is correct, and nothing here reproduces the paper's results. A line beginning “fixture:” under a sample that ran gives the arguments Syntology's fixture called the sample with, each as name:type/rank/element kind (? where none was recorded), and the type of what the fixture's run returned, exactly as the harness recorded them; the arguments and the returned type were recorded by two separate runs of the same fixture. No output value is shown, and a returned type says nothing about whether the output was right. “Level recorded” gives the UTC date when Syntology's graph wrote the sample's verification level, not when the code executed; where the level was written before Syntology's graph kept that date, the row says “level recorded, date not recorded”. “Honoured” and “violated” refer to a contract Syntology drafted from the code itself. “Our draft was wrong” and “the contract check could not run” are failures of Syntology's instrument, not of the code: the sample ran on Syntology's synthesized input, and then the drafted contract did not describe it, or the check itself raised an error (RAISES), lacked a dependency (DEP_MISSING) or ran out of time (TIMEOUT); hover the label for which. A metamorphic tier, where shown, is evidence of the output's shape, not that the sample implements the method.

Each sample ends with its code_sha256. A sample's code_sha256 is Syntology's identity for that exact code. It is printed in full (16 hexadecimal characters for every sample in this build), and get_code(code_sha256="…") takes it exactly as printed. An agent calls it on Syntology's MCP service (how to connect); click an identity to copy that call.

Repository labels, per sample. official repository: The archive marks this repository official for the paper. named in the paper: The archive records that the paper mentions this repository; it is not marked official. community (archive-listed): In the archive's code links for this paper, not marked official and not recorded as mentioned in the paper. found in paper text by Syntology: 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. Samples are grouped by repository: a repository marked official first, then by how directly the repository is tied to the paper, then in the archive's order; never by how many ran. Licence labels name the repository's licence as recorded at harvest. “Pointer only” means Syntology does not serve that copy's text, for one of six reasons: no licence file was found; the licence was not identified; the licence is recorded as permissive but that copy's record is not marked cleared; the licence is outside the permissive list Syntology serves text under (MIT, Apache-2.0, BSD and similar); this paper's copy was not located, so its licence is not recorded; or this paper's copy is cleared, but the stored copy of the same code that Syntology serves is not. Some licences outside that list permit redistribution, such as WTFPL, and GPL-3.0 under its conditions; they are simply not on the list. Hover a licence label for the reason. File links open the file on GitHub at the default branch, which may have changed since the harvest.

A function or class name links to its page when the name occurs in the harvested code of at least 5 papers; the name is a string, not an identity.

lions-epfl/charmer official repository 31 samples, 23 ran (13 instrument-failed) · 8 unverified
CrossEntropyLoss_negative charmer.py ran MIT (permissive) · 6a768c7d89fa45a6 · report fixture: signature logits:Tensor/2/float, true_classes:Tensor/1/int · returned Tensor · level recorded 2026-09-24
all_strings_within_edit1 charmer.py ran · our draft was wrong MIT (permissive) · b87dfb2a13227d05 · report fixture: signature sequence:str/0/str, bases:str/0/str · returned list · level recorded 2026-09-22
all_strings_within_editx charmer.py ran · our draft was wrong MIT (permissive) · 232efbd63c8df619 · report fixture: signature sequence:str/0/str, bases:str/0/str, edit_distance:int/0/int · returned set · level recorded 2026-09-22
backtrace charmer.py ran; the contract check could not run (RAISES) MIT (permissive) · 96a7b58d20a2005f · report fixture: signature first:str/0/str, second:str/0/str, matrix:list/1/list · returned tuple · level recorded 2026-09-22
compute_centroid_distances baseline/roben/agglom_clusters.py ran · honoured contract fingerprinted MIT (permissive) · 92dfc8c506947d1c · report fixture: signature X:ndarray/2/float, mu:ndarray/2/float · returned ndarray · level recorded 2026-09-24
concat_prompt_question_label charmer.py ran · our draft was wrong MIT (permissive) · 7a4e3ab49ac3c159 · report fixture: signature T_promptbegin:dict/0/?, T_questions:dict/0/?, T_promptend:dict/0/?, T_label:dict/0/? · returned dict · level recorded 2026-09-22
ctx_noparamgrad charmer.py ran MIT (permissive) · 4e36262a542c16fb · report fixture: returned ctx_noparamgrad · level recorded 2026-09-22
generate_all_sentences charmer.py ran · our draft was wrong fingerprinted MIT (permissive) · 7b06bf5b0955fdac · report fixture: signature S:str/0/str, V:list/1/int, subset_z:NoneType/0/?, k:int/0/int, alternative:int/0/int · returned list · level recorded 2026-09-22
generate_all_sentences_at_z charmer.py ran · our draft was wrong MIT (permissive) · b82a34d25dd75308 · report fixture: signature S:str/0/str, z:int/0/int, V:list/1/int, k:int/0/int, alternative:int/0/int · returned list · level recorded 2026-09-22
generate_sentence charmer.py ran · our draft was wrong MIT (permissive) · 7f56e8436fe9b621 · report fixture: signature S:str/0/str, z:int/0/int, u:int/0/int, V:list/1/int, k:int/0/int, alternative:NoneType/0/? · returned str · level recorded 2026-09-22
get_class_num dataloader.py ran · honoured contract fingerprinted MIT (permissive) · 904d63da437332d4 · report fixture: signature dataset_name:str/0/str · returned int · level recorded 2026-09-24
get_goals_and_targets utils_jailbreak.py ran · our draft was wrong MIT (permissive) · 55554ebb64125749 · report fixture: signature params:Params/0/? · returned tuple · level recorded 2026-09-24
get_module_training_state context.py ran · our draft was wrong MIT (permissive) · 28413f6ed2f091a9 · report fixture: signature module:callable/0/? · returned dict · level recorded 2026-09-24
get_param_grad_state charmer.py ran · our draft was wrong MIT (permissive) · 41eb44fa1a771cd1 · report fixture: signature module:callable/0/? · returned dict · level recorded 2026-09-22
get_task_type dataloader.py ran · violated contract MIT (permissive) · 75df2d0c0186e517 · report fixture: signature dataset_name:str/0/str · returned bool · level recorded 2026-09-24
get_vocabulary in 11 papers utils.py ran MIT (permissive) · 593614c84d8b37c2 · report fixture: signature dataset:dict/0/?, dataset_name:str/0/str · returned list · level recorded 2026-09-24
llm_input_processor charmer.py ran MIT (permissive) · 456e7d855a79eb1e · report fixture: returned llm_input_processor · level recorded 2026-09-22
margin_loss_llm charmer.py ran MIT (permissive) · 28879c448358c558 · report fixture: returned margin_loss_llm · level recorded 2026-09-22
margin_loss_lm charmer.py ran · our draft was wrong fingerprinted MIT (permissive) · 7eae0eca72c38f98 · report fixture: signature logits:Tensor/2/float, true_class:int/0/int · returned Tensor · level recorded 2026-09-22
margin_loss_lm_batched charmer.py ran MIT (permissive) · f38056701ce7c0b6 · report fixture: signature logits:Tensor/2/float, true_classes:Tensor/1/int · returned Tensor · level recorded 2026-09-24
target_loss in 5 papers utils_jailbreak.py ran MIT (permissive) · 11a3348eb0e4ba60 · report fixture: signature logits:Tensor/3/float, ids:Tensor/2/int, target_slice:slice/0/? · returned Tensor · level recorded 2026-09-24
update_mu baseline/roben/agglom_clusters.py ran; the contract check could not run (RAISES) fingerprinted MIT (permissive) · 12efa3ddd30cf7c2 · report fixture: signature C:ndarray/2/float, f:ndarray/1/float · returned ndarray · level recorded 2026-09-24
word_edit_distance charmer.py ran · our draft was wrong fingerprinted MIT (permissive) · f0c1289fd2565267 · report fixture: signature x:str/0/str, y:str/0/str · returned tuple · level recorded 2026-09-22
Charmer charmer.py unverified MIT (permissive) · 59ab5813bed83cb2 · report
combined_emb_input utils.py unverified MIT (permissive) · a299d58465c7bf59 · report
get_llm_embeddings charmer.py unverified MIT (permissive) · 8ae9f9ad369707a9 · report
get_llm_encoder charmer.py unverified MIT (permissive) · 6297161e122df004 · report
load_attack_dataset dataloader.py unverified MIT (permissive) · 7f504481b40a6171 · report
resize_embeddings charmer.py unverified MIT (permissive) · cb53a78f621babdb · report
set_param_grad_off charmer.py unverified MIT (permissive) · e75d10fb3cfbb60c · report
set_param_grad_state charmer.py unverified MIT (permissive) · ea9f4cd8cc668205 · report

Method as code Syntology

What Syntology's graph records as this paper's claimed method, lifted into code: each assembly names its bricks (functions and classes), each with its own role and its own verification level. An assembly is not a check that the code implements the method correctly or reproduces the paper's results. Levels are each brick's own verification level in Syntology's graph (level 0: no recorded run), listed brick by brick and never combined into one level; “ran” means the brick executed on a synthesized input, not that its output is correct, and the instrument figure counts failures of Syntology's instrument, not of the code. Where a model read the code against the paper's claim, each assembly's label verdict is one model's judgement, kept in this page's JSON twin; it is not calibrated, so no card prints it.

1 assembly on this paper: 1 harvested, 0 generated.

harvested Assembly for Charmer (graph method: Charmer). 21 bricks; individually on a fixture: 15 ran (of which 3 constructed an object rather than computing a result; 5 with no instrument failure: 0 honoured, 0 violated, 5 with no contract checked; 10 where Syntology's instrument failed) · 6 unverified. The assembly has not been run end-to-end.

  1. Charmer entry · level 0 · unverified · 59ab5813bed83cb2
  2. CrossEntropyLoss_negative helper · level 1 · ran · 6a768c7d89fa45a6
  3. all_strings_within_editx helper · level 1 · ran · our draft was wrong · 232efbd63c8df619
  4. concat_prompt_question_label helper · level 1 · ran · our draft was wrong · 7a4e3ab49ac3c159
  5. ctx_noparamgrad helper · level 1 · ran · 4e36262a542c16fb
  6. generate_all_sentences helper · level 1 · ran · our draft was wrong · 7b06bf5b0955fdac
  7. get_llm_embeddings helper · level 0 · unverified · 8ae9f9ad369707a9
  8. get_llm_encoder helper · level 0 · unverified · 6297161e122df004
  9. llm_input_processor helper · level 1 · ran · 456e7d855a79eb1e
  10. margin_loss_llm helper · level 1 · ran · 28879c448358c558
  11. margin_loss_lm_batched helper · level 1 · ran · f38056701ce7c0b6
  12. resize_embeddings helper · level 0 · unverified · cb53a78f621babdb
  13. all_strings_within_edit1 helper · level 1 · ran · our draft was wrong · b87dfb2a13227d05
  14. get_param_grad_state helper · level 1 · ran · our draft was wrong · 41eb44fa1a771cd1
  15. set_param_grad_off helper · level 0 · unverified · e75d10fb3cfbb60c
  16. set_param_grad_state helper · level 0 · unverified · ea9f4cd8cc668205
  17. generate_all_sentences_at_z helper · level 1 · ran · our draft was wrong · b82a34d25dd75308
  18. margin_loss_lm helper · level 1 · ran · our draft was wrong · 7eae0eca72c38f98
  19. backtrace helper · level 1 · ran; the contract check could not run (RAISES) · 96a7b58d20a2005f
  20. word_edit_distance helper · level 1 · ran · our draft was wrong · f0c1289fd2565267
  21. generate_sentence helper · level 1 · ran · our draft was wrong · 7f56e8436fe9b621

Each brick ends with its code_sha256, Syntology's identity for that exact code; click one to copy the get_code call that returns it. For agents, on Syntology's MCP service: assembly(arxiv_id="2405.04346") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

Adversarial AttackSentence

1 archive task tag without a task page not shown.

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

AdamAttentionAttention DropoutBERTDense ConnectionsDropoutLayer NormalizationLinear LayerLinear Warmup With Linear DecayMulti-Head AttentionResidual ConnectionSoftmaxUSEWeight DecayWordPiece

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections