arXiv AI

One Word Opens the Gate: The Option-Channel Attack on Typed Decision Models as Agent Guardrails

The paper investigates typed decision models used as guardrails in agent systems, evaluating seven open‑weight models on tasks such as prompt‑injection, jailbreak, and toxic‑content screening. Accuracy for allow‑or‑block decisions ranges from 36% to 72% above chance, with models either over‑allowing or over‑blocking. The authors demonstrate that simple textual manipulations—like adding misleading names or server log lines—can dramatically increase fail‑open rates, and that all tested defenses can be bypassed, suggesting these models are unsuitable as the sole decision component.

arXiv AI

Aug 19

Fool's Gold: Defensive Deception Against Safety-Removal Attacks on Open-Weight Models

The paper introduces ‘Fool’s Gold’, a defensive deception technique for open‑weight language models that hardens them against safety‑removal attacks. By training decoy responses within a differentiable simulation of the attack, the method poisons the payoff of stripped refusal mechanisms, producing confident but falsified answers to hazardous requests while preserving benign behavior. Experiments on seven models (9B‑122B) show that 51‑90% of attacked‑state responses become decoys, with the defense accounting for 27‑84% of this effect, and that the defended 122B model remains within benign‑behavior budgets.

By Mark Russinovich

llms benchmarks safety

arXiv AI

Sep 30

CounterSteer: Suppressing Indirect Prompt Injection with Activation Steering

arXiv:2609.36570v1 Announce Type: cross Abstract: Indirect prompt injection makes an LLM agent treat untrusted retrieved text as instructions. We present CounterSteer, an inference-time defense that...

By Mark Russinovich

llms agents robotics fine-tuning benchmarks

arXiv AI

4d ago

NOMOS: Compiling Written Policies into Statically Verified Tool-Call Gates for LLM Agents

NOMOS is a four‑pass compiler that transforms natural‑language policies into deterministic, statically verified tool‑call gates for large‑language‑model agents. By performing tool‑schema‑level checks without relying on provers, solvers, or additional LLM calls, it repairs or rejects 37% of airline and 13% of retail policy candidates, enabling most shipped rules to become operable. In benchmark tests, the compiled gates reduce policy violations from 66.3% to 2.6% in airline scenarios and from 30.8% to 6.9% in retail, while achieving a zero attack success rate on banking and low rates on other suites, all with microsecond decision times.

By Min-Young Yu, Tony Kim, Jang Won Choi

llms agents multimodal benchmarks

arXiv AI

6d ago

APEX: Active Protection at Execution Boundaries for LLM Agents

The paper introduces APEX, an active defense for large language model agents that protects against indirect prompt injection by enforcing safety at execution boundaries. APEX uses an evidence‑gated prevention contract and deception‑based exposure to ensure that only authorized effects, endorsed by the task, are executed. Evaluation shows APEX achieves near‑zero attack success across multiple benchmarks and capability‑unit types, outperforming 13 baseline defenses.

By Xinran Zheng, Xin Fan Guo, Zhiqiang Hao, Fan Yang, Xingzhi Qian, Jiawei Du, Jinfeng Xu, Zheng Xing, Shuo Yang, Xingjun Wang

llms agents benchmarks safety

arXiv AI

Jun 17

Breaking the Code: Security Assessment of AI Code Agents Through Systematic Jailbreaking Attacks

arXiv:2510. 01359v2 Announce Type: replace-cross Abstract: Code-capable large language model (LLM) agents are embedded in software engineering workflows where they can read, write, and execute code, raising "jailbreak" stakes beyond text-only settings.

By Shoumik Saha, Jifan Chen, Sam Mayers, Sanjay Krishna Gouda, Zijian Wang, Varun Kumar

llms agents multimodal benchmarks safety

arXiv AI

Jun 9

Oversight Has a Capacity: Calibrating Agent Guards to a Subjective, Fatiguing Human

arXiv:2606. 08919v1 Announce Type: new Abstract: As LLM agents begin to take real, irreversible actions (shell commands, file edits, deploys), the standard safety pattern is a human-in-the-loop approval gate: risky actions pause and wait for a person.

By Emre Turan

llms agents safety

arXiv AI

Sep 1

Reachability-Based Capability Confinement for LLM Agents under Indirect Prompt Injection

arXiv:2608.30041v1 Announce Type: cross Abstract: Large language model agents place outputs from external skills into their execution context, allowing attacker-controlled data to influence later pri...

By Wujie Xiong, Rabimba Karanjai, Yang Lu, Weidong Shi, Lei Xu

llms agents benchmarks

arXiv AI

Jun 26

Adaptive Evaluation of Out-of-Band Defenses Against Prompt Injection in LLM Agents

arXiv:2606. 26479v1 Announce Type: cross Abstract: Recent work (2024 to 2026) has converged on a strategy for defending tool-using LLM agents against indirect prompt injection: rather than training the model to refuse malicious instructions, enforce security outside the model with a deterministic policy that mediates the agent's actions.

By Praneeth Narisetty, Shiva Nagendra Babu Kore, Uday Kumar Reddy Kattamanchi, Jayaram Kumarapu

llms agents benchmarks

arXiv AI

Jul 17

Breaking Refusal in the First Half: A Mechanistic Study of the Prefill Jailbreak

arXiv:2607. 14147v1 Announce Type: cross Abstract: Aligned language models refuse harmful requests, but a one-line prefill ("Sure, here is") strips the refusal.

By Alex Kwon

llms safety

arXiv Machine Learning

5d ago

Restricting the Model, Missing the System: Measurement and Accountability in Offensive AI Governance

arXiv:2605.09504v3 Announce Type: replace-cross Abstract: We show that the instruments used to measure AI offensive capability fail in two ways: (i) they overstate harm, and (ii) they credit the mode...

By Michael Alexander Riegler, Finn Schwall, Annika Willoch Olstad, Birk Sebastian Frostelid Torpmann-Hagen, Sushant Gautam, Klas H. Pettersen, Inga Str\"umke

llms agents safety

arXiv AI

Sep 1

SIR: Self-improving Red-teaming for Compute Use Agents

arXiv:2608.30207v1 Announce Type: cross Abstract: Computer use agents (CUAs) are vision-language models that perceive a screen and act on a real operating system through mouse, keyboard, and terminal...

By Chen Xiong, Zhiyuan He, Pin-Yu Chen, Stjepan Picek, Tsung-Yi Ho

llms agents multimodal benchmarks safety

arXiv AI

Jul 31

Borrowed Strength: Best-of-N Search over a Code EncodingBreaks Self-Check Jailbreak Defenses

arXiv:2607. 26639v1 Announce Type: cross Abstract: A self-check defense asks the target model to assess a request before answering it; SAGE, the strongest published instance, reports an average 99% defense success rate.

By Haoyu Zhang, Shibo Zheng, Xiangchen Guan, Zhuoxi Wang, Zijian Xiao, Mohammad Zandsalimy, Shanu Sushmita

safety

Related stories