Pith Number

pith:GUD5X3KT

pith:2026:GUD5X3KTZYOHJ3YCGCV4XF3LGQ

not attested not anchored not stored refs resolved

Inference-Time Machine Unlearning via Gated Activation Redirection

Christian Mattjie, Flavio du Pin Calmon, Joana Pasquali, Jo\~ao Vitor Boer Abitante, Kristen K. Arguello, Lucas S. Kupssinsk\"u, Ot\'avio Parraga, Ramiro N. Barros, Rodrigo C. Barros, Vin\'icius Conte Turani

A gated input-dependent rotation in the residual stream enables inference-time unlearning of specific data in LLMs without altering weights.

arxiv:2605.12765 v1 · 2026-05-12 · cs.LG

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{GUD5X3KTZYOHJ3YCGCV4XF3LGQ}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

GUARD-IT matches or exceeds 12 gradient-based baselines across three model scales, while being the only method to simultaneously preserve utility, suppress memorization, and avoid catastrophic collapse across all settings.

C2weakest assumption

That an input-dependent norm-preserving rotation in the residual stream can selectively remove the influence of a forget set without introducing unintended changes to model behavior on unrelated inputs.

C3one line summary

GUARD-IT performs machine unlearning in LLMs via inference-time gated activation redirection, matching or exceeding gradient-based baselines on TOFU and MUSE while preserving utility and working under quantization.

References

59 extracted · 59 resolved · 4 Pith anchors

[1] TOFU: A Task of Fictitious Unlearning for LLMs · arXiv:2401.06121

[2] Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks , url =

[3] Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019 2019

[4] Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , year =

[5] Smith and Chiyuan Zhang , booktitle= 2025

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-05-18T03:09:48.303388Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

3507dbed53ce1c74ef0230abcb976b343c0a43b6b181873b53bc79e4fbdab568

Aliases

arxiv: 2605.12765 · arxiv_version: 2605.12765v1 · doi: 10.48550/arxiv.2605.12765 · pith_short_12: GUD5X3KTZYOH · pith_short_16: GUD5X3KTZYOHJ3YC · pith_short_8: GUD5X3KT

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/GUD5X3KTZYOHJ3YCGCV4XF3LGQ \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 3507dbed53ce1c74ef0230abcb976b343c0a43b6b181873b53bc79e4fbdab568

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "98b122ffa4d5a003357308d804bbb6194227a3a3366125f7e6df692a29b1174a",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by/4.0/",
    "primary_cat": "cs.LG",
    "submitted_at": "2026-05-12T21:26:25Z",
    "title_canon_sha256": "f7ac0148a6ac44443446671da58bd82b96960f87e1e7ee0de219e17cb0d846d7"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2605.12765",
    "kind": "arxiv",
    "version": 1
  }
}