Visual perturbations alter VLA actions without changing text rationale
TNAS-guided attacks induce collisions in 52.6% of driving scenes while keeping rationales word-for-word identical.
Stories per day · 30 days
Relatedagentic1availability1cryptography1geocloak1geolocation1hash-function1inference1lvlm1preimage1privacy1reasoning1sha2561text-action1vla1vlm1
10 stories, newest first · one card per story
Stories tagged #adversarial
Hush-Up attack halts distributed LLM inference
A single malicious worker forces premature LLM termination with small output perturbations that evade local activation-based verification.
Practical preimages for 19- and 20-step SHA-256 with standard padding
One CPU core solves arbitrary 19-step padded targets in median 19.5 seconds.
GeoCloak: An Agentic Framework for Semantics-Preserving Geolocation Privacy via Peripheral Clue Manipulation
GeoCloak cuts LVLM city-level geolocation accuracy from 41.2% to 9.9% while preserving image meaning.
Epistemic Typing as a PostgreSQL Table Access Method: Adversarial Conflict Resolution Under Confidence Forgery and Sybil Coordination
KNDB, a PostgreSQL 18 access method, resists confidence forgery and Sybil writes using engine-assigned epistemic types.
LENS: The Sum Is Worse Than the Parts for Set-Level Poisoning in Retrieval-Augmented Generation
LENS poisons RAG by composing document sets whose attack succeeds only when the full set is retrieved.
AERIAL: Adversarial Evaluation of Robustness in Accuracy-Preserving Low-Precision EEG Decoders
A research paper on low-precision EEG decoders examines adversarial robustness across compression techniques.
The Tokens Remember: When Tokenization Bypasses Knowledge Editing and Unlearning
Toketive shows alternative tokenizations can bypass LLM knowledge edits and recover suppressed answers.
Decoding Guardrails: XAI-Guided Perturbation Analysis of Prompt Injection Detection
A paper analyzes XAI techniques to evaluate Prompt Guard 2's prompt injection detection, finding that explanation methods may reduce the cost of adversarial bypasses.
Adversarial Testing of Automated Program Repair Agents for Security Vulnerabilities
SWEADV benchmark shows adversarial issue descriptions make LLM program-repair agents write insecure fixes in 51.7% of cases, evading most detection tools.