9 papers
An Actionable Diagnosis of Multilingual, Multi-Agent Planning Failures
Vikas Pahuja, Jonathan Brokman, Omer Hofman +6
Multilingual multi-agent systems exhibit substantial degradation beyond English, yet prior work rarely identifies how task-critical information is lost when user requests are conve…
Inference-Time Backdoors via Chat Templates: From LLM Supply Chains to Agentic System Compromise
Ariel Fogel, Omer Hofman, Eilon Cohen +1
Open-weight language models are increasingly used in production settings, raising new security challenges. One prominent threat is backdoor attacks, in which adversaries embed hidd…
When Scanners Lie: Evaluator Instability in LLM Red-Teaming
Lidor Erez, Omer Hofman, Tamir Nizri +1
Automated LLM vulnerability scanners are increasingly used to assess security risks by measuring different attack type success rates (ASR). Yet the validity of these measurements h…
Identifying Memorization of Diffusion Models through -Laplace Analysis: Estimators, Bounds and Applications
Jonathan Brokman, Itay Gershon, Amit Giloni +4
Diffusion models, today's leading image generative models, estimate the score function, i.e. the gradient of the log probability of (perturbed) data samples, without direct access…
MAPS: A Multilingual Benchmark for Agent Performance and Security
Omer Hofman, Jonathan Brokman, Oren Rachmil +7
Agentic AI systems, which build on Large Language Models (LLMs) and interact with tools and memory, have rapidly advanced in capability and scope. Yet, since LLMs have been shown t…
Training-Free Policy Violation Detection via Activation-Space Whitening in LLMs
Oren Rachmil, Avishag Shapira, Roy Betser +5
As organizations increasingly deploy LLMs in sensitive domains such as legal, financial, and medical settings, ensuring alignment with internal organizational policies has become a…