Paper Citation Record · LEDGER

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior

As of 23 August 2026, this Paper Citation Record lists 25 of 25 outbound references and 0 inbound Pith citation observations for arXiv:2606.18322.

A citation records a reference. It does not transfer a finding from one paper to another.

pith.paper-citation-record.v1
2606.18322 v1

Coverage vector

measured 25 of 25 reference resolution

Typed states for the displayed outbound observations.

Source: paper_references, paper_reference_links, observed 2026-06-27T01:33:34.151529Z

measured 25 of 25 standing notices

One-hop event checks from named stored sources.

Source: scholarly_work_events, retraction_status_cache, observed 2026-08-23T06:30:58.430688+00:00

measured 0 of 0 inbound itemization

Pith citing papers itemized under the disclosed page cap.

Source: paper_references, paper_reference_links

measured 0 of 1 external citation measurements

A source-named dated measurement, never combined with another source.

Source: cited_works

Reference resolution

25 of 25 outbound references displayed

verified exact0
verified fuzzy0
unresolved24
parse uncertain0
malformed identifier1
metadata mismatch0

External citation measurements

No source-named external measurement is stored.

Outbound references

Observation caf6ce93-f816-4a34-aa15-78cb94015289 · outbound

This paper cites Saes are good for steering – if you select the right features, 2025.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Saes are good for steering – if you select the right features, 2025

Reference 1

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 65837b4b-7ab8-475d-ad1d-927a6b908ba4 · outbound

This paper cites Obfuscated activations bypass llm latent-space defenses, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Obfuscated activations bypass llm latent-space defenses, 2024

Reference 2

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 43cb661a-327d-4873-a040-6832f11694b5 · outbound

This paper cites an unresolved cited work.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Unresolved cited work

Reference 3

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation ab8b8c45-c1bf-4c30-9c5b-dbf63c562204 · outbound

This paper cites A is for absorption: Studying feature splitting and absorption in sparse autoen- coders, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior A is for absorption: Studying feature splitting and absorption in sparse autoen- coders, 2024

Reference 4

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation c6d959ae-5260-4e5f-a782-75c3f136683e · outbound

This paper cites Feature hedging: Correlated features break narrow sparse autoencoders, 2025.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Feature hedging: Correlated features break narrow sparse autoencoders, 2025

Reference 5

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 61c10199-7a50-48be-9201-a5a74f66fd47 · outbound

This paper cites Sparse autoencoders find highly interpretable features in language models, 2023.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Sparse autoencoders find highly interpretable features in language models, 2023

Reference 6

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1dc74c2d-5094-454f-b2fb-2996cfa433d9 · outbound

This paper cites Toy models of superposition, 2022.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Toy models of superposition, 2022

Reference 7

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation dabeb51e-9f13-4a0c-bb37-489628c6c39b · outbound

This paper cites Applying sparse autoencoders to unlearn knowledge in language models, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Applying sparse autoencoders to unlearn knowledge in language models, 2024

Reference 8

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 3726d579-3d46-4270-b4c5-9e88193b8d96 · outbound

This paper cites Scaling and evaluating sparse autoencoders, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Scaling and evaluating sparse autoencoders, 2024

Reference 9

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7a9eb5ee-c4fe-4582-b82d-149665cbd45a · outbound

This paper cites Saebench: A comprehensive benchmark for sparse autoencoders in language model interpretability, 2025.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Saebench: A comprehensive benchmark for sparse autoencoders in language model interpretability, 2025

Reference 10

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation d9f31c5e-f84f-4d6d-b6ae-0c31647d8419 · outbound

This paper cites Inference- time intervention: Eliciting truthful answers from a language model, 2023.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Inference- time intervention: Eliciting truthful answers from a language model, 2023

Reference 11

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation f6e8b154-5f52-4289-af91-4810a60de126 · outbound

This paper cites Michaud, David D.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Michaud, David D

Reference 12

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation de97e201-70de-4796-94af-d4993b2f5893 · outbound

This paper cites Gemma scope: Open sparse autoencoders everywhere all at once on gemma 2, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Gemma scope: Open sparse autoencoders everywhere all at once on gemma 2, 2024

Reference 13

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 332b05da-158c-4375-a29d-87e77a7081e3 · outbound

This paper cites Towards principled evaluations of sparse autoencoders for interpretability and control, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Towards principled evaluations of sparse autoencoders for interpretability and control, 2024

Reference 14

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 84d26531-f337-488d-8a7f-374acffe87c1 · outbound

This paper cites Michaud, Yonatan Belinkov, David Bau, and Aaron Mueller.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Michaud, Yonatan Belinkov, David Bau, and Aaron Mueller

Reference 15

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 4ef34202-2bff-4557-92ef-20083012ea0e · outbound

This paper cites Steering language model refusal with sparse autoencoders, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Steering language model refusal with sparse autoencoders, 2024

Reference 16

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 06c30836-41d5-4885-ad3b-0241903a110b · outbound

This paper cites Steering llama 2 via contrastive activation addition, 2024.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Steering llama 2 via contrastive activation addition, 2024

Reference 17

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fcd7346b-5a98-4581-a3b0-1a977fb84336 · outbound

This paper cites Multilingual Knowledge Graph Completion with Self-Supervised Adaptive Graph Alignment.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Multilingual Knowledge Graph Completion with Self-Supervised Adaptive Graph Alignment

Reference 18

Resolution: malformed identifier; doi_truncated, observed 2026-06-27T01:40:21.283652Z

Source-reported events for the cited work

No event found in the named queried sources as of 2026-08-23T06:30:58.430688+00:00.

Observation 08c74058-d575-42cc-aa99-a09be1779c98 · outbound

This paper cites Latentguard: Controllable latent steering for robust refusal of attacks and reliable response generation, 2025.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Latentguard: Controllable latent steering for robust refusal of attacks and reliable response generation, 2025

Reference 19

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 7fa0c743-411b-4090-915f-f817367e8922 · outbound

This paper cites Turner, Callum McDougall, Monte MacDiarmid, Alex Tamkin, Esin Durmus, Tristan Hume, Francesco Mosconi, C.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Turner, Callum McDougall, Monte MacDiarmid, Alex Tamkin, Esin Durmus, Tristan Hume, Francesco Mosconi, C

Reference 20

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation e4834726-ae9c-4ff2-b8f2-4173ddb66951 · outbound

This paper cites Vazquez, Ulisse Mini, and Monte MacDiarmid.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Vazquez, Ulisse Mini, and Monte MacDiarmid

Reference 21

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation fe20f161-9b67-423a-9126-2bc297b57693 · outbound

This paper cites Interpretability in the wild: A circuit for indirect object identification in gpt-2 small, 2022.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Interpretability in the wild: A circuit for indirect object identification in gpt-2 small, 2022

Reference 22

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 02ed2835-17ae-46cd-8a51-514a23781301 · outbound

This paper cites Understanding refusal in language models with sparse autoencoders, 2025.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Understanding refusal in language models with sparse autoencoders, 2025

Reference 23

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation 1e20e7a2-e569-4671-b53e-6c1383bc45da · outbound

This paper cites Byun, Zifan Wang, Alex Mallen, Steven Basart, Sanmi Koyejo, Dawn Song, Matt Fredrikson, J.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior Byun, Zifan Wang, Alex Mallen, Steven Basart, Sanmi Koyejo, Dawn Song, Matt Fredrikson, J

Reference 24

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Observation abfa22c9-ea1a-41f7-952d-b9e0129048ca · outbound

This paper cites I cannot.

SAE Interventions are Unreliable: Post-Intervention Recovery of Suppressed Behavior I cannot

Reference 25

Resolution: unresolved; no resolver link, observed 2026-06-27T01:33:34.151529Z

Source-reported events for the cited work

Unavailable: canonical work link unavailable.

Pith citing papers

No inbound Pith citation observations are available.