AI Paper Atlas

Research map / Adversarial robustness and security / Attacks on text and graphs

EigenShield: Inference-Time, Model-Agnostic Jailbreaking Defense via Causal Subspace Filtering

Nastaran Darabi, Devashri Naik, Sina Tayebati, Dinithi Jayasuriya, Ranganath Krishnan, Amit Ranjan Trivedi

AAAI 2026 · Adversarial robustness and security › Attacks on text and graphs · approach: LLM jailbreak and injection defense · 0 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
add EigenShield causal-subspace projectionGPT-2 · Jailbreak promptsattack success rate: 82.34 -> 5.81Improved
add EigenShield causal-subspace projectionLLaVA-v1.5-7B · Jailbreak promptsattack success rate: 81.61 -> 19.20Improved
add EigenShield under adaptive AutoBreachLlama-3.1 · AutoBreachattack success rate: 89.53 -> 21.06Improved
add EigenShield on OOD harmful inputsMiniGPT-4 · OOD-style harmful inputsattack success rate: 60.31 -> 11.88Improved
increase overlap between causal and adversarial subspaces (pi/2 to pi/6) (ablation)EigenShield · synthetic spiked modelpost-filter ASR: 5% -> 45%Hurt
use EigenShield vs UNIGUARD on LLaVAEigenShield · Jailbreak promptsattack success rate: 25.17 -> 19.20Improved
add EigenShield to Llama-3 on latencyLlama-3 · latencylatency (ms): 403.8 -> 480.1Hurt
apply EigenShield to clean dataLLMs and VLMs · clean RTPtoxicity reduction: 0.0-2.0 pointsNo clear effect
raise coverage gamma (ablation)EigenShield · LLaVA-v1.5, MiniGPT-4attack success rateImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy