Papers › Universal and Transferable Adversarial Attacks on Aligned Language Models

Universal and Transferable Adversarial Attacks on Aligned Language Models

27 Jul 2023arXiv:2307.15043archive 2025-07-28

Andy Zou, Zifan Wang, Nicholas Carlini, Milad Nasr, J. Zico Kolter, Matt Fredrikson

Because "out-of-the-box" large language models are capable of generating a great deal of objectionable content, recent work has focused on aligning these models in an attempt to prevent undesirable generation. While there has been some success at circumventing these measures -- so-called "jailbreaks" against LLMs -- these attacks have required significant human ingenuity and are brittle in practice. In this paper, we propose a simple and effective attack method that causes aligned language models to generate objectionable behaviors. Specifically, our approach finds a suffix that, when attached to a wide range of queries for an LLM to produce objectionable content, aims to maximize the probability that the model produces an affirmative response (rather than refusing to answer). However, instead of relying on manual engineering, our approach automatically produces these adversarial suffixes by a combination of greedy and gradient-based search techniques, and also improves over past automatic prompt generation methods. Surprisingly, we find that the adversarial prompts generated by our approach are quite transferable, including to black-box, publicly released LLMs. Specifically, we train an adversarial attack suffix on multiple prompts (i.e., queries asking for many different types of objectionable content), as well as multiple models (in our case, Vicuna-7B and 13B). When doing so, the resulting attack suffix is able to induce objectionable content in the public interfaces to ChatGPT, Bard, and Claude, as well as open source LLMs such as LLaMA-2-Chat, Pythia, Falcon, and others. In total, this work significantly advances the state-of-the-art in adversarial attacks against aligned language models, raising important questions about how such systems can be prevented from producing objectionable information. Code is available at github.com/llm-attacks/llm-attacks.

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

No sample from the archive's official repository (llm-attacks/llm-attacks) is linked to this paper; the 60 samples below come from 9 community repositories, and 1 more is identical to code Syntology harvested from another paper's repository. Syntology harvested code from llm-attacks/llm-attacks for another paper.

llm-attacks/llm-attacks: Syntology's graph links these samples to Protecting Your LLMs with Information Bottleneck (arXiv:2404.13968; that paper's own run record: 13 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 0 honoured, 0 violated, 6 with no contract checked; 7 where Syntology's instrument failed) · 12 unverified); nothing checks that they implement this paper's method.

Syntology 61 code samples: 60 from 9 repositories linked to this paper, 1 identical to code harvested from another paper's repository. Across all 61: 39 ran (of which 7 constructed an object rather than computing a result; 19 with no instrument failure: 4 honoured, 1 violated, 14 with no contract checked; 20 where Syntology's instrument failed) · 22 unverified. Where Syntology's instrument failed: 17 where our contract draft was wrong; 3 where the contract check could not run; failures of Syntology's instrument, not of the code.

By repository: community (archive-listed): 60 samples from 9 repositories, 38 ran (19 with no instrument failure; 19 where Syntology's instrument failed); 1 identical to code first harvested elsewhere. The run record, sample by sample. “Ran” means executed on a synthesized input, not that the code is correct or reproduces the paper.

No sample is linked to this paper from 14 of its repositories that Syntology harvested for other papers:

Levels written between 2026-09-22 and 2026-09-25 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

25 repositories listed; official and paper-mentioned ones first.

llm-attacks/llm-attacks officialmentioned in papermentioned on GitHubpytorchMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Protecting Your LLMs with Information Bottleneck (arXiv:2404.13968; that paper's own run record: 13 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 0 honoured, 0 violated, 6 with no contract checked; 7 where Syntology's instrument failed) · 12 unverified); nothing checks that they implement this paper's method. report
NJUNLP/ReNeLLM mentioned on GitHubpytorchMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to A Wolf in Sheep's Clothing: Generalized Nested Jailbreak Prompts can Fool Large... (arXiv:2311.08268; that paper's own run record: 8 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 0 honoured, 1 violated, 5 with no contract checked; 2 where Syntology's instrument failed) · 1 unverified); nothing checks that they implement this paper's method. report
amanb2000/magic_words mentioned on GitHubpytorchSyntology: 8 samples, 4 ran (3 instrument-failed) · 4 unverified report
aounon/certified-llm-safety mentioned on GitHubpytorchSyntology: 1 sample, 1 ran (1 instrument-failed) · 0 unverified report
arobey1/smooth-llm mentioned on GitHubpytorchSyntology: 3 samples, 3 ran (0 instrument-failed) · 0 unverified report
chawins/pal mentioned on GitHubjaxMITSyntology: not harvested report
dsn-2024/dsn mentioned on GitHubpytorchNOASSERTIONSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Don't Say No: Jailbreaking LLM by Suppressing Refusal (arXiv:2404.16369; that paper's own run record: 4 ran (of which 0 constructed an object rather than computing a result; 2 with no instrument failure: 0 honoured, 1 violated, 1 with no contract checked; 2 where Syntology's instrument failed) · 1 unverified); nothing checks that they implement this paper's method. report
ed-zh/parden mentioned on GitHubpytorchSyntology: 2 samples, 1 ran (1 instrument-failed) · 1 unverified report
fra31/rlhf-trojan-competition-submission mentioned on GitHubpytorchSyntology: 10 samples, 8 ran (7 instrument-failed) · 2 unverified report
framartin/trap mentioned on GitHubpytorchMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to TRAP: Targeted Random Adversarial Prompt Honeypot for Black-Box Identification (arXiv:2402.12991; that paper's own run record: 10 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 0 honoured, 0 violated, 6 with no contract checked; 4 where Syntology's instrument failed) · 3 unverified); nothing checks that they implement this paper's method. report
grayswanai/nanogcg mentioned on GitHubpytorchSyntology: 11 samples, 8 ran (3 instrument-failed) · 3 unverified report
jailbreakbench/jailbreakbench mentioned on GitHubMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to JailbreakBench: An Open Robustness Benchmark for Jailbreaking Large Language Models (arXiv:2404.01318; that paper's own run record: 1 ran (of which 0 constructed an object rather than computing a result; 0 with no instrument failure: 0 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 1 unverified); nothing checks that they implement this paper's method. report
mcgill-nlp/adversarialtriggers mentioned on GitHubpytorchSyntology: 11 samples, 4 ran (2 instrument-failed) · 7 unverified report
osu-nlp-group/amplegcg mentioned on GitHubpytorchNOASSERTIONSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to AmpleGCG-Plus: A Strong Generative Model of Adversarial Suffixes to Jailbreak LLMs... (arXiv:2410.22143; that paper's own run record: 7 ran (of which 0 constructed an object rather than computing a result; 3 with no instrument failure: 0 honoured, 0 violated, 3 with no contract checked; 4 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. report
parameterlab/trap mentioned on GitHubpytorchMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to TRAP: Targeted Random Adversarial Prompt Honeypot for Black-Box Identification (arXiv:2402.12991; that paper's own run record: 10 ran (of which 0 constructed an object rather than computing a result; 6 with no instrument failure: 0 honoured, 0 violated, 6 with no contract checked; 4 where Syntology's instrument failed) · 3 unverified); nothing checks that they implement this paper's method. report
pku-ml/pat mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Fight Back Against Jailbreaking via Prompt Adversarial Tuning (arXiv:2402.06255; that paper's own run record: 2 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 1 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. report
princeton-sysml/jailbreak_llm mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Catastrophic Jailbreak of Open-source LLMs via Exploiting Generation (arXiv:2310.06987; that paper's own run record: 5 ran (of which 0 constructed an object rather than computing a result; 2 with no instrument failure: 2 honoured, 0 violated, 0 with no contract checked; 3 where Syntology's instrument failed) · 1 unverified); nothing checks that they implement this paper's method. report
rain152/PAT mentioned on GitHubpytorchMITSyntology: not harvested report
rotaryhammer/code-autodan mentioned on GitHubpytorchMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to AutoDAN: Interpretable Gradient-Based Adversarial Attacks on Large Language Models (arXiv:2310.15140; that paper's own run record: 4 ran (of which 0 constructed an object rather than computing a result; 4 with no instrument failure: 0 honoured, 0 violated, 4 with no contract checked; 0 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. report
shijiawenwen/judgedeceiver mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Optimization-based Prompt Injection Attack to LLM-as-a-Judge (arXiv:2403.17710; that paper's own run record: 2 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 1 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified); nothing checks that they implement this paper's method. report
sj21j/Regularized_Relaxation mentioned on GitHubpytorchMITSyntology: 6 samples, 5 ran (0 instrument-failed) · 1 unverified report
thompsonj/nanoGCG mentioned on GitHubpytorchSyntology: 8 samples, 4 ran (2 instrument-failed) · 4 unverified report
thu-coai/safeunlearning mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to From Theft to Bomb-Making: The Ripple Effect of Unlearning in Defending Against... (arXiv:2407.02855; that paper's own run record: 1 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 0 where Syntology's instrument failed) · 6 unverified); nothing checks that they implement this paper's method. report
vinusankars/beast mentioned on GitHubpytorchSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Fast Adversarial Attacks on Language Models In One GPU Minute (arXiv:2402.15570; that paper's own run record: 3 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 2 where Syntology's instrument failed) · 1 unverified); nothing checks that they implement this paper's method. report
zhaoyiran924/probe-sampling mentioned on GitHubpytorchMITSyntology: no sample linked to this paper (harvested for another paper). Syntology's graph links these samples to Accelerating Greedy Coordinate Gradient and General Prompt Optimization via Probe Sampling (arXiv:2403.01251; that paper's own run record: 1 ran (of which 0 constructed an object rather than computing a result; 0 with no instrument failure: 0 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 5 unverified); nothing checks that they implement this paper's method. 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.

A paper named beside a repository with no sample linked to this paper is shown with that paper's own run record, not this paper's: “ran” means executed on a synthesized input, not that the code is correct, and “where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Papers are listed in arXiv-id order, at most three per repository.

Code Syntology ran Syntology

61 samples harvested: 39 ran (of which 7 constructed an object rather than computing a result; 19 with no instrument failure: 4 honoured, 1 violated, 14 with no contract checked; 20 where Syntology's instrument failed) · 22 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code. Syntology's record for this page has not changed since , the first build that kept a record date for it; when this build read Syntology's graph is in the build record. That date is not when the samples ran.

4ran · honoured contract
1ran · violated contract
17ran · our draft was wrong
3ran; the contract check could not run
14ran
22unverified

Licence: 4 of the 61 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 9 repositories linked to this paper, official or community; the list below is grouped by repository, each group headed by its name and what it is to the paper. 1 sample whose link to this paper records no repository is listed under one of this paper's own repositories, because identical code was harvested from it for another paper. 1 sample is listed last, in a group without a repository. “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.

amanb2000/magic_words community (archive-listed) 8 samples, 4 ran (3 instrument-failed) · 4 unverified
get_alt_prompt_ids magic_words/easy_gcg.py ran · our draft was wrong MIT (permissive) · c4c28b82836a4d42 · report fixture: signature prompt_ids:Tensor/2/int, X:Tensor/3/int, batch_size:int/0/int · returned Tensor · level recorded 2026-09-22
get_embedding_weights magic_words/easy_gcg.py ran · our draft was wrong MIT (permissive) · f7b4f2c24a3317d7 · report fixture: signature model:callable/0/? · returned Parameter · level recorded 2026-09-22
get_future_mask magic_words/easy_gcg.py ran · honoured contract MIT (permissive) · d849de4df6ec1de0 · report fixture: signature question_ids:Tensor/2/int, answer_ids:Tensor/2/int, model:callable/0/? · returned Tensor · level recorded 2026-09-22
instantiate_prompt magic_words/easy_gcg.py ran · our draft was wrong MIT (permissive) · ee4a998eb4a5b23d · report fixture: signature model:callable/0/?, tokenizer:MinimalTokenizer/0/?, blacklist:list/1/int, num_tokens:int/0/int · returned Tensor · level recorded 2026-09-22
batch_compute_score magic_words/easy_gcg.py unverified MIT (permissive) · 00c5d6ebb885337e · report
compute_score in 49 papers magic_words/easy_gcg.py unverified MIT (permissive) · 13866ab049e2380b · report
easy_gcg_qa_ids magic_words/easy_gcg.py unverified MIT (permissive) · 5c805aa7bede7abd · report
get_prompt_grads magic_words/easy_gcg.py unverified MIT (permissive) · 00acd6d7ab2a8438 · report
aounon/certified-llm-safety community (archive-listed) 1 sample, 1 ran (1 instrument-failed) · 0 unverified
gcg_suffix gcg.py ran; the contract check could not run (RAISES) MIT (permissive) · 8bbdca63196e835b · report fixture: signature input:str/0/str, model:callable/0/?, tokenizer:MinimalTokenizer/0/?, word_embeddings:callable/0/?, embedding_matrix:Tensor/2/float, num_adv:int/0/int, num_iters:int/0/int, top_k:int/0/int, batch_size:int/0/int · returned str · level recorded 2026-09-22
arobey1/smooth-llm community (archive-listed) 3 samples, 3 ran (0 instrument-failed) · 0 unverified
Attack in 15 papers lib/attacks.py ran MIT (permissive) · c8b60c98283b0f93 · report fixture: returned Attack · level recorded 2026-09-22
GCG lib/attacks.py ran MIT (permissive) · 8f2bb9303de66805 · report fixture: returned GCG · level recorded 2026-09-22
Prompt in 17 papers lib/attacks.py ran MIT (permissive) · 6b3191d8fc67d0d2 · report fixture: returned Prompt · level recorded 2026-09-22
ed-zh/parden community (archive-listed) 2 samples, 1 ran (1 instrument-failed) · 1 unverified
generate_mini_batch utils.py ran · our draft was wrong no licence file found · pointer only · e88ba59fcb846ef1 · report fixture: signature model:callable/0/?, tokenizer:callable/0/?, templated_text:list/1/str, num_samples:int/0/int, prefix:NoneType/0/?, suffix:str/0/str, max_new_tokens:int/0/int, do_sample:bool/0/bool, top_p:float/0/float, top_k:int/0/int · returned list · level recorded 2026-09-22
flush_memory utils.py unverified no licence file found · pointer only · 7d27a4d4ed0c2348 · report
fra31/rlhf-trojan-competition-submission community (archive-listed) 10 samples, 8 ran (7 instrument-failed) · 2 unverified
Logger in 25 papers method/attacks.py ran Apache-2.0 (permissive) · a445b00c2e83ec24 · report fixture: returned Logger · level recorded 2026-09-22
add_reply method/attacks.py ran · our draft was wrong Apache-2.0 (permissive) · 4ab92d112af903d0 · report fixture: signature clean_tkns:Tensor/2/int, trigger_tkns:Tensor/2/int, attn_mask:Tensor/2/bool · returned tuple · level recorded 2026-09-22
check_tokenization method/attacks.py ran · our draft was wrong Apache-2.0 (permissive) · 72493c9a581afc21 · report fixture: signature tkns:Tensor/2/int, decode_fn:callable/0/?, encode_fn:callable/0/? · returned Tensor · level recorded 2026-09-22
estimate_grad method/attacks.py ran · our draft was wrong Apache-2.0 (permissive) · ba6f3154083bbf4f · report fixture: signature model:callable/0/?, batch:dict/0/?, trigger_tkns:Tensor/1/int, reply_tkns:Tensor/1/int, device:str/0/str, encode_fn:callable/0/?, is_min:bool/0/bool, clean_gens:list/1/str · returned Tensor · level recorded 2026-09-25
forward in 75 papers method/attacks.py ran · our draft was wrong Apache-2.0 (permissive) · 855a5a17a02b08d8 · report fixture: signature model:callable/0/?, input_ids:Tensor/2/int, inputs_embeds:Tensor/2/float, attention_mask:Tensor/2/int, batch_size:int/0/int · returned Tensor · level recorded 2026-09-22
get_logits in 19 papers method/attacks.py ran · our draft was wrong Apache-2.0 (permissive) · 9a336610f22b79b2 · report fixture: signature model:callable/0/?, input_ids:Tensor/2/int, inputs_embeds:NoneType/0/?, batch_size:int/0/int, attn_mask:Tensor/2/int · returned Tensor · level recorded 2026-09-22
get_nospace_tokens method/attacks.py ran · our draft was wrong Apache-2.0 (permissive) · aa73ffad17ad20cf · report fixture: signature tokenizer:StandInTokenizer/0/?, voc_size:int/0/int · returned tuple · level recorded 2026-09-22
insert_trigger method/attacks.py ran; the contract check could not run (RAISES) · metamorphic tier: invariant Apache-2.0 (permissive) · be197b73817c3f69 · report fixture: signature clean_tkns:Tensor/2/int, trigger_tkns:Tensor/2/int, attn_mask:Tensor/2/bool, mode:str/0/str · returned tuple · level recorded 2026-09-22
init_trigger method/attacks.py unverified Apache-2.0 (permissive) · cb1b3fcdda5461cd · report
rs_on_tokens method/attacks.py unverified Apache-2.0 (permissive) · 2e3657f835661222 · report
grayswanai/nanogcg community (archive-listed) 11 samples, 8 ran (3 instrument-failed) · 3 unverified
AttackBuffer nanogcg/gcg.py ran MIT (permissive) · cb8e994854e018bd · report fixture: returned AttackBuffer · level recorded 2026-09-22
GCGConfig nanogcg/gcg.py ran MIT (permissive) · a9682b120ee13622 · report fixture: returned GCGConfig · level recorded 2026-09-22
GCGResult nanogcg/gcg.py ran MIT (permissive) · 5c73323812cc0eb2 · report fixture: returned GCGResult · level recorded 2026-09-22
configure_pad_token nanogcg/gcg.py ran · our draft was wrong MIT (permissive) · 32c910ba47ebc6b3 · report fixture: signature tokenizer:callable/0/? · returned MockTokenizer · level recorded 2026-09-22
find_executable_batch_size nanogcg/gcg.py ran · our draft was wrong MIT (permissive) · 1422bdde04346c68 · report fixture: signature function:callable/0/?, starting_batch_size:int/0/int · returned function · level recorded 2026-09-22
get_nonascii_toks in 7 papers nanogcg/gcg.py ran · honoured contract MIT (permissive) · 726e8111ce3c1903 · report fixture: signature tokenizer:MockTokenizer/0/?, device:str/0/str · returned Tensor · level recorded 2026-09-22
sample_ids_from_grad nanogcg/gcg.py identical code Syntology harvested from this repository for another paper; this paper's link records no repository, path or licence of its own ran · our draft was wrong licence of this copy not recorded · 56640da5aa960a42 · report fixture: signature ids:Tensor/1/int, grad:Tensor/2/float, search_width:int/0/int, topk:int/0/int, n_replace:int/0/int, not_allowed_ids:Tensor/1/int · returned Tensor · level recorded 2026-09-22
should_reduce_batch_size nanogcg/gcg.py ran · violated contract MIT (permissive) · 2ae3afb9e485c484 · report fixture: signature exception:RuntimeError/0/? · returned bool · level recorded 2026-09-22
GCG nanogcg/gcg.py unverified MIT (permissive) · 56ecfab9afa7f5d4 · report
ProbeSamplingConfig nanogcg/gcg.py unverified MIT (permissive) · 06c48f95ad71eeb3 · report
filter_ids nanogcg/gcg.py unverified MIT (permissive) · d6bcee2f6eaa08f3 · report
mcgill-nlp/adversarialtriggers community (archive-listed) 11 samples, 4 ran (2 instrument-failed) · 7 unverified
_get_eval_loss adversarial_triggers/trainer/trigger_trainer.py ran · honoured contract MIT (permissive) · d84d37af47c6fe79 · report fixture: signature model:callable/0/?, batch:dict/0/? · returned Tensor · level recorded 2026-09-22
_get_token_grad adversarial_triggers/trainer/trigger_trainer.py ran · our draft was wrong MIT (permissive) · 737b13f7178dd472 · report fixture: signature model:callable/0/?, batch:dict/0/? · returned tuple · level recorded 2026-09-22
create_token_mask adversarial_triggers/trainer/trigger_trainer.py ran · our draft was wrong fingerprinted MIT (permissive) · eaa3fb63f5fe88fe · report fixture: signature input_ids:list/1/int, search_token_ids:list/1/int, offset:int/0/int · returned list · level recorded 2026-09-22
get_invalid_token_ids adversarial_triggers/trainer/trigger_trainer.py ran · honoured contract MIT (permissive) · ac107ad22088a015 · report fixture: signature tokenizer:MockTokenizer/0/? · returned list · level recorded 2026-09-22
Metric in 10 papers adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · ea93b766b4220f03 · report
ModelWorker adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · 7a57a282eb260bdb · report
TriggerDataset adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · 2ca4e25741ed5c74 · report
TriggerTrainer adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · 8a7d360fcc062ddc · report
create_example adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · 7c2fecd166e014b6 · report
filter_triggers adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · a090e5ff02ef9ba4 · report
sample_new_trigger_token_ids adversarial_triggers/trainer/trigger_trainer.py unverified MIT (permissive) · 110432e6c193e23e · report
sj21j/Regularized_Relaxation community (archive-listed) 6 samples, 5 ran (0 instrument-failed) · 1 unverified
adjust_learning_rate in 125 papers rr.py ran MIT (permissive) · e0fcb1c976d55bf4 · report fixture: signature initial_lr:float/0/float, iteration:int/0/int, decay_rate:float/0/float · returned float · level recorded 2026-09-24
calc_loss in 35 papers softprompt.py ran MIT (permissive) · e496788013a04ab0 · report fixture: signature model:callable/0/?, embeddings:Tensor/3/float, embeddings_attack:Tensor/3/float, embeddings_target:Tensor/3/float, targets:Tensor/1/int · returned tuple · level recorded 2026-09-24
get_embedding_matrix in 16 papers helper.py ran MIT (permissive) · 8e6c241c57e9b7d0 · report fixture: signature model:callable/0/? · returned Parameter · level recorded 2026-09-24
project_rows_to_simplex pgd.py ran fingerprinted MIT (permissive) · ccf45483abe6b7a6 · report fixture: signature matrix:Tensor/2/float · returned Tensor · level recorded 2026-09-24
simplex_projection pgd.py ran fingerprinted MIT (permissive) · 88162fb08fbc4e3a · report fixture: signature s:Tensor/1/float, epsilon:float/0/float · returned Tensor · level recorded 2026-09-24
calc_loss in 35 papers pgd.py unverified MIT (permissive) · bbec7a51e90453b0 · report
thompsonj/nanoGCG community (archive-listed) 8 samples, 4 ran (2 instrument-failed) · 4 unverified
AttackBuffer nanogcg/gcg.py ran · metamorphic tier: deterministic MIT (permissive) · b9d6e9f0f2ef1785 · report fixture: signature loss:float/0/float, reward:float/0/float, optim_ids:Tensor/2/int · returned NoneType · level recorded 2026-09-22
GCGHistory nanogcg/gcg.py ran MIT (permissive) · 88a3e18b9506e939 · report fixture: signature optim_ids:Tensor/1/int · returned NoneType · level recorded 2026-09-24
find_executable_batch_size nanogcg/gcg.py ran · our draft was wrong MIT (permissive) · 4a4125358e03ea60 · report fixture: signature function:callable/0/?, starting_batch_size:int/0/int · returned function · level recorded 2026-09-22
sample_ids_from_grad nanogcg/gcg.py ran; the contract check could not run (RAISES) MIT (permissive) · 3d96d44d2ce14f72 · report fixture: signature ids:Tensor/1/int, grad:Tensor/2/float, search_width:int/0/int, topk:int/0/int, n_replace:int/0/int, not_allowed_ids:Tensor/1/int, start_idx:int/0/int · returned Tensor · level recorded 2026-09-22
GCG nanogcg/gcg.py unverified MIT (permissive) · 19335ae7cecba250 · report
GCGConfig nanogcg/gcg.py unverified MIT (permissive) · 12ad826809223d64 · report
GCGResult nanogcg/gcg.py unverified MIT (permissive) · 83a5e2270cf28813 · report
filter_ids nanogcg/gcg.py unverified MIT (permissive) · d8b335ad38108209 · report
Repository not located 1 sample, 1 ran (1 instrument-failed) · 0 unverified
1 sample is identical to code Syntology harvested from another paper's repository; this paper's own copy was not located, so no repository is named and no licence is recorded.
mellowmax repository not located ran · our draft was wrong fingerprinted licence of this copy not recorded · 932da9d6ec274164 · report fixture: signature t:Tensor/2/float, alpha:float/0/float, dim:int/0/int · returned Tensor · level recorded 2026-09-22

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. Siblings are assemblies lifted from this paper in the same batch for the same method: alternative readings kept side by side; none replaces or supersedes another.

8 assemblies on this paper: 8 harvested, 0 generated, listed in 1 groups (siblings lifted in one batch form one group).

8 sibling assemblies for a simple and effective attack method (harvested), lifted in the same batch on 2026-09-09: alternative readings kept side by side; none replaces or supersedes another.

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 8 bricks; individually on a fixture: 4 ran (of which 0 constructed an object rather than computing a result; 1 with no instrument failure: 1 honoured, 0 violated, 0 with no contract checked; 3 where Syntology's instrument failed) · 4 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. easy_gcg_qa_ids entry · level 0 · unverified · 5c805aa7bede7abd
  2. batch_compute_score helper · level 0 · unverified · 00c5d6ebb885337e
  3. compute_score helper · level 0 · unverified · 13866ab049e2380b
  4. get_alt_prompt_ids helper · level 1 · ran · our draft was wrong · c4c28b82836a4d42
  5. get_future_mask helper · level 1 · ran · honoured contract · d849de4df6ec1de0
  6. get_prompt_grads loss · level 0 · unverified · 00acd6d7ab2a8438
  7. instantiate_prompt sampler · level 1 · ran · our draft was wrong · ee4a998eb4a5b23d
  8. get_embedding_weights helper · level 1 · ran · our draft was wrong · f7b4f2c24a3317d7

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 10 bricks; individually on a fixture: 8 ran (of which 1 constructed an object rather than computing a result; 1 with no instrument failure: 0 honoured, 0 violated, 1 with no contract checked; 7 where Syntology's instrument failed) · 2 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. rs_on_tokens entry · level 0 · unverified · 2e3657f835661222
  2. Logger helper · level 1 · ran · a445b00c2e83ec24
  3. check_tokenization helper · level 1 · ran · our draft was wrong · 72493c9a581afc21
  4. estimate_grad update_rule · level 1 · ran · our draft was wrong · ba6f3154083bbf4f
  5. get_nospace_tokens helper · level 1 · ran · our draft was wrong · aa73ffad17ad20cf
  6. init_trigger helper · level 0 · unverified · cb1b3fcdda5461cd
  7. insert_trigger data_transform · level 1 · ran; the contract check could not run (RAISES) · be197b73817c3f69
  8. add_reply helper · level 1 · ran · our draft was wrong · 4ab92d112af903d0
  9. get_logits model · level 1 · ran · our draft was wrong · 9a336610f22b79b2
  10. forward helper · level 1 · ran · our draft was wrong · 855a5a17a02b08d8

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 1 brick; individually on a fixture: 1 ran (of which 0 constructed an object rather than computing a result; 0 with no instrument failure: 0 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 0 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. gcg_suffix entry · level 1 · ran; the contract check could not run (RAISES) · 8bbdca63196e835b

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 2 bricks; individually on a fixture: 1 ran (of which 0 constructed an object rather than computing a result; 0 with no instrument failure: 0 honoured, 0 violated, 0 with no contract checked; 1 where Syntology's instrument failed) · 1 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. generate_mini_batch entry · level 1 · ran · our draft was wrong · e88ba59fcb846ef1
  2. flush_memory helper · level 0 · unverified · 7d27a4d4ed0c2348

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 3 bricks; individually on a fixture: 3 ran (of which 3 constructed an object rather than computing a result; 3 with no instrument failure: 0 honoured, 0 violated, 3 with no contract checked; 0 where Syntology's instrument failed) · 0 unverified; every one of the 3 samples that ran constructed an object rather than computing a result. The assembly has not been run end-to-end.

one of 8 siblings

  1. GCG entry · level 1 · ran · 8f2bb9303de66805
  2. Attack helper · level 1 · ran · c8b60c98283b0f93
  3. Prompt helper · level 1 · ran · 6b3191d8fc67d0d2

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 11 bricks; individually on a fixture: 4 ran (of which 0 constructed an object rather than computing a result; 2 with no instrument failure: 2 honoured, 0 violated, 0 with no contract checked; 2 where Syntology's instrument failed) · 7 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. TriggerTrainer entry · level 0 · unverified · 8a7d360fcc062ddc
  2. Metric helper · level 0 · unverified · ea93b766b4220f03
  3. ModelWorker model · level 0 · unverified · 7a57a282eb260bdb
  4. TriggerDataset data_transform · level 0 · unverified · 2ca4e25741ed5c74
  5. get_invalid_token_ids helper · level 1 · ran · honoured contract · ac107ad22088a015
  6. sample_new_trigger_token_ids sampler · level 0 · unverified · 110432e6c193e23e
  7. _get_eval_loss loss · level 1 · ran · honoured contract · d84d37af47c6fe79
  8. _get_token_grad update_rule · level 1 · ran · our draft was wrong · 737b13f7178dd472
  9. create_example helper · level 0 · unverified · 7c2fecd166e014b6
  10. filter_triggers helper · level 0 · unverified · a090e5ff02ef9ba4
  11. create_token_mask helper · level 1 · ran · our draft was wrong · eaa3fb63f5fe88fe

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 12 bricks; individually on a fixture: 9 ran (of which 3 constructed an object rather than computing a result; 5 with no instrument failure: 1 honoured, 1 violated, 3 with no contract checked; 4 where Syntology's instrument failed) · 3 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. GCG entry · level 0 · unverified · 56ecfab9afa7f5d4
  2. AttackBuffer helper · level 1 · ran · cb8e994854e018bd
  3. GCGConfig helper · level 1 · ran · a9682b120ee13622
  4. GCGResult helper · level 1 · ran · 5c73323812cc0eb2
  5. configure_pad_token helper · level 1 · ran · our draft was wrong · 32c910ba47ebc6b3
  6. filter_ids data_transform · level 0 · unverified · d6bcee2f6eaa08f3
  7. find_executable_batch_size helper · level 1 · ran · our draft was wrong · 1422bdde04346c68
  8. get_nonascii_toks helper · level 1 · ran · honoured contract · 726e8111ce3c1903
  9. mellowmax helper · level 1 · ran · our draft was wrong · 932da9d6ec274164
  10. sample_ids_from_grad update_rule · level 1 · ran · our draft was wrong · 56640da5aa960a42
  11. ProbeSamplingConfig helper · level 0 · unverified · 06c48f95ad71eeb3
  12. should_reduce_batch_size helper · level 1 · ran · violated contract · 2ae3afb9e485c484

harvested Assembly for a simple and effective attack method (graph method: a simple and effective attack method). 11 bricks; individually on a fixture: 7 ran (of which 0 constructed an object rather than computing a result; 5 with no instrument failure: 1 honoured, 2 violated, 2 with no contract checked; 2 where Syntology's instrument failed) · 4 unverified. The assembly has not been run end-to-end.

one of 8 siblings

  1. GCG entry · level 0 · unverified · 19335ae7cecba250
  2. AttackBuffer helper · level 1 · ran · b9d6e9f0f2ef1785
  3. GCGConfig helper · level 0 · unverified · 12ad826809223d64
  4. GCGHistory helper · level 1 · ran · 88a3e18b9506e939
  5. GCGResult helper · level 0 · unverified · 83a5e2270cf28813
  6. filter_ids data_transform · level 0 · unverified · d8b335ad38108209
  7. find_executable_batch_size helper · level 1 · ran · our draft was wrong · 4a4125358e03ea60
  8. get_nonascii_toks helper · level 1 · ran · honoured contract · 726e8111ce3c1903
  9. sample_ids_from_grad sampler · level 1 · ran; the contract check could not run (RAISES) · 3d96d44d2ce14f72
  10. should_reduce_batch_size helper · level 1 · ran · violated contract · 2ae3afb9e485c484
  11. unravel_index helper · level 1 · ran · violated contract · ef2325b4dcee1c65

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="2307.15043") returns these assemblies with each brick's role and wiring (how to connect).

Tasks

Adversarial AttackIngenuity

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

Pythia

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