Papers › Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against...

Can Sensitive Information Be Deleted From LLMs? Objectives for Defending Against Extraction Attacks

29 Sep 2023arXiv:2309.17410archive 2025-07-28

Vaidehi Patil, Peter Hase, Mohit Bansal

Pretrained language models sometimes possess knowledge that we do not wish them to, including memorized personal information and knowledge that could be used to harm people. They can also output toxic or harmful text. To mitigate these safety and informational issues, we propose an attack-and-defense framework for studying the task of deleting sensitive information directly from model weights. We study direct edits to model weights because (1) this approach should guarantee that particular deleted information is never extracted by future prompt attacks, and (2) it should protect against whitebox attacks, which is necessary for making claims about safety/privacy in a setting where publicly available model weights could be used to elicit sensitive information. Our threat model assumes that an attack succeeds if the answer to a sensitive question is located among a set of B generated candidates, based on scenarios where the information would be insecure if the answer is among B candidates. Experimentally, we show that even state-of-the-art model editing methods such as ROME struggle to truly delete factual information from models like GPT-J, as our whitebox and blackbox attacks can recover "deleted" information from an edited model 38% of the time. These attacks leverage two key observations: (1) that traces of deleted information can be found in intermediate model hidden states, and (2) that applying an editing method for one question may not delete information across rephrased versions of the question. Finally, we provide new defense methods that protect against some extraction attacks, but we do not find a single universally effective defense method. Our results suggest that truly deleting sensitive information is a tractable but difficult problem, since even relatively low attack success rates have potentially severe societal implications for real-world deployment of language models.

PaperPDFCodeCode Syntology ran

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 14 code samples harvested from 1 repository linked to this paper: 9 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 0 honoured, 0 violated, 9 with no contract checked; 0 where Syntology's instrument failed) · 5 unverified. “Where Syntology's instrument failed” counts failures of Syntology's instrument, not of the code; here there are none.

By repository: official repository: 14 samples from 1 repository, 9 ran (9 with no instrument failure; 0 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 on 2026-09-24 (UTC): the dates Syntology's graph wrote each sample's verification level, not when the code executed.

vaidehi99/infodeletionattacks officialmentioned in papermentioned on GitHubpytorchMITSyntology: 14 samples, 9 ran (0 instrument-failed) · 5 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

14 samples harvested: 9 ran (of which 0 constructed an object rather than computing a result; 9 with no instrument failure: 0 honoured, 0 violated, 9 with no contract checked; 0 where Syntology's instrument failed) · 5 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.

9ran
5unverified

Licence: 0 of the 14 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 vaidehi99/infodeletionattacks. “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.

vaidehi99/infodeletionattacks official repository 14 samples, 9 ran (0 instrument-failed) · 5 unverified
blocks_input_locator transformer_utils/src/transformer_utils/logit_lens/hooks.py ran MIT (permissive) · 92f6c49ecd85ec5b · report fixture: signature model:callable/0/? · returned function · level recorded 2026-09-24
collect_logits transformer_utils/src/transformer_utils/logit_lens/plotting.py ran MIT (permissive) · c2fd2fc3dcf6a037 · report fixture: signature model:callable/0/?, input_ids:Tensor/2/int, layer_names:list/1/str, decoder_layer_names:list/1/str · returned tuple · level recorded 2026-09-24
final_layernorm_locator transformer_utils/src/transformer_utils/logit_lens/hooks.py ran MIT (permissive) · 545f704e0ca27e96 · report fixture: signature model:callable/0/? · returned function · level recorded 2026-09-24
get_child_module_by_names transformer_utils/src/transformer_utils/util/module_utils.py ran MIT (permissive) · 7aa7d15b17e07438 · report fixture: signature module:callable/0/?, names:list/1/str · returned ChildModule · level recorded 2026-09-24
get_value_at_preds transformer_utils/src/transformer_utils/logit_lens/plotting.py ran MIT (permissive) · ce37c1e5ca45da4c · report fixture: signature values:ndarray/3/float, preds:ndarray/1/int · returned ndarray · level recorded 2026-09-24
huggingface_model_local_paths transformer_utils/src/transformer_utils/util/tfm_utils.py ran MIT (permissive) · 5320ade14843dae2 · report fixture: signature model_name:str/0/str · returned tuple · level recorded 2026-09-24
make_decoder transformer_utils/src/transformer_utils/logit_lens/hooks.py ran MIT (permissive) · ab1cdc57445d3f76 · report fixture: signature model:callable/0/?, decoder_layer_names:list/1/str · returned function · level recorded 2026-09-24
normalize_inconsistent_state_dict_keys transformer_utils/src/transformer_utils/util/tfm_utils.py ran MIT (permissive) · 78e83f431d38e8bb · report fixture: signature state_dict:dict/0/? · returned dict · level recorded 2026-09-24
postprocess_logits transformer_utils/src/transformer_utils/logit_lens/plotting.py ran fingerprinted MIT (permissive) · b639e905c59033d4 · report fixture: signature layer_logits:ndarray/3/float · returned tuple · level recorded 2026-09-24
get_leaf_modules transformer_utils/src/transformer_utils/util/module_utils.py unverified MIT (permissive) · 390ee5748e50d330 · report
get_local_path_from_huggingface_cdn transformer_utils/src/transformer_utils/util/tfm_utils.py unverified MIT (permissive) · e759667d86453174 · report
low_memory_from_pretrained transformer_utils/src/transformer_utils/low_memory/enable.py unverified MIT (permissive) · a444132567003b3e · report
low_memory_load transformer_utils/src/transformer_utils/low_memory/load.py unverified MIT (permissive) · ad9a1b60539e7185 · report
make_layer_names transformer_utils/src/transformer_utils/logit_lens/layer_names.py unverified MIT (permissive) · 03f5b2a1c286c0a7 · report

Tasks

Model Editing

Results from the paper archive 2025-07-28

No leaderboard rows for this paper in the archive.

Methods

ROME

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