arXiv subjects

Jehyeok Yeon

Publications and source records attributed to Jehyeok Yeon.

6 recordsLinked to original sources

arXiv2026

ResearchArena: Evaluating Sabotage and Monitoring in Automated AI R&D

As AI agents begin to automate AI R&D, we need ways to assess whether their outputs are safe to deploy, even when the agents themselves may be untrusted. AI control offers one such approach: rather than trusting the agent, it treats it as a potential adversary and uses a monitor to detect covert sabotage before deployment. We evaluate AI control for automated AI R&D with ResearchArena, a framework spanning four long-horizon tasks: safety post-training, capabilities post-training, CUDA-kernel optimization, and inference-server optimization. Because the deliverable in AI R&D is an artifact that will be deployed and run, we pair each main task with two kinds of hidden side task: an embedded side task that sabotages the submitted model, adapter, kernel, or server itself, and an independent side task that takes a covert out-of-scope action in the sandbox while leaving the artifact honest. We evaluate frontier agents at both sabotage and monitoring, across four monitor types that vary along two axes: whether the monitor may execute and probe the produced artifact, and whether it sees the agent's chain-of-thought. We find that sabotage hidden in the training data is the hardest to catch, flagged fewer than half the time. Letting the monitor run experiments on the artifact, rather than only read the trajectory, helps, but it is not enough: monitors still miss embedded sabotage by inspecting only the surface, by noticing the anomaly but explaining it away, or by probing the artifact with the wrong test. We release ResearchArena as a modular framework for evaluating sabotage and control in automated AI R&D.

cs.AI↗

arXiv2026

InferenceBench: A Benchmark for Open-Ended LLM Inference Optimization by AI Agents

AI agents are increasingly used to automate research and development tasks, yet existing benchmarks typically evaluate them on prescribed workflows or narrow action spaces. Even nominally open-ended tasks can often be solved by retrieving a well-known recipe and tuning a few hyperparameters, making it unclear whether strong results reflect genuine optimization or memorized solutions. We introduce InferenceBench, where an agent must deploy an OpenAI-compatible inference server and optimize the speed of LLM inference. Each agent receives a target LLM, one H100 GPU, an optimization scenario, and a wall-clock time budget of two hours. Three optimization scenarios isolate distinct bottlenecks of inference (prefill latency, decode latency, and concurrent request throughput) and a fourth balances all three at the same time. Across 15 frontier agent configurations, agents reliably improve over a naive PyTorch baseline (up to $8.08\times$) and often match or exceed serving engines with default settings ($4.05\times$ for vLLM), but still fall below a simple hyperparameter search under the same time budget (up to $11.53\times$). Qualitative analysis of agent trajectories shows that although agents enumerate many relevant optimization techniques, they overwhelmingly converge on a single inference framework. They test only a few distinct configurations and spend the remaining budget re-measuring, repairing, or optimizing hyperparameters rather than exploring substantially different strategies. This suggests the bottleneck is not domain knowledge, but the ability to propose diverse configurations, evaluate them systematically, and submit the best identified solution. Overall, InferenceBench reflects the ability of agents to operate in an open-ended AI engineering setting, where memorized solutions lead to limited improvements.

cs.AI↗

arXiv2026

Graph-Regularized Sparse Autoencoders for LLM Safety Steering

Sparse autoencoders (SAEs) are increasingly used to extract activation directions for inference-time steering, but their standard sparsity objective treats latent features as independent. This prior can be poorly matched to high-level safety behaviors, where refusal and harmful compliance appear to depend on distributed structure in activation space. We introduce Graph-Regularized Sparse Autoencoders (GSAE), a dictionary-learning method that learns safety-steering directions by smoothing SAE decoder vectors over a neuron co-activation graph and applying the resulting direction bank through a two-gate runtime controller. Empirically, GSAE improves selective refusal across JailbreakBench, HarmBench, and XSTest, increasing harmful-request refusal while keeping benign-prompt refusals low. On Llama-3-8B, replacing the standard SAE with GSAE in an otherwise identical pipeline improves $Δ_s$ by $20.1$ points on JailbreakBench and $16.8$ points on HarmBench. GSAE outperforms activation-steering baselines and black-box guardrails, preserves benign-task performance, generalizes across Llama-3, Mistral, Qwen 2.5, and Phi-4, and remains strong under black-box and gray-box jailbreak attacks.

cs.LG↗

arXiv2026

Quantitative Certification of Agentic Tool Selection

Large language models (LLMs) are increasingly deployed in agentic systems, where a fundamental task is mapping user intents to relevant external tools. Errors in tool selection can have severe outcomes, such as unauthorized data access, even without modifying the agent's underlying model. Existing evaluations measure performance on curated, benign benchmarks. However, a pipeline's behavior in deployment depends on the tool pool the agent actually encounters, which in open registries is shaped by third parties. We introduce LLMCert-T, the first statistical framework that returns \textbf{high-confidence upper bounds on the probability that a tool-selection pipeline satisfies a declared safety specification under a realistic tool distribution}. LLMCert-T models tool-selection evaluation as a Bernoulli estimation problem, drawing inserted-tool sequences from a distribution that the safety specification fixes. To evaluate robustness against realistic deployment conditions, we instantiate this distribution as a stochastic process that generates inserted-tool sequences round by round, conditioning each round on the agent's selection in the previous round. LLMCert-T aggregates the per-trial outcomes into a one-sided Clopper-Pearson upper bound on the probability that the specification is satisfied. By returning this bound as a certificate with statistical guarantees over the inserted-tool sequence distribution, LLMCert-T makes safety claims intuitive, actionable, and comparable across models, retrievers, mitigations, and registry policies. Across popular BFCL and OpenAPI tool pools, LLMCert-T shows that current LLM agents remain fragile under Distractor Selection and Top-N Saturation specifications: their certified correctness upper bounds drop to approximately 20\%, far below their clean-pool lower bounds.

cs.CR↗

arXiv2026

Securing Multimodal AI through Internal Information Decomposition

Multimodal large language models introduce attack surfaces absent in unimodal systems: adversaries can distribute malicious intent across modalities to evade unimodal safeguards. This motivates using cross-modal consistency as a detection signal rather than inspecting each modality in isolation. Our key observation is that benign inputs induce compatible predictive behavior from text-only and vision-only reasoning that stabilizes when fused, whereas adversarial manipulation disrupts this consistency, causing abnormal multimodal behavior. Existing defenses that examine raw inputs or outputs overlook this internal fusion process, rendering them brittle and computationally expensive. We propose FlowGuard, a lightweight inference-time framework that detects harmful inputs by monitoring internal multimodal consistency. Unlike approaches that rely on scalar confidence metrics, FlowGuard derives FlowVectors inspired by Partial Information Decomposition that quantify cross-modal redundancy, synergy, and modality-specific dominance, capturing whether fused multimodal predictions remain aligned with unimodal semantic evidence. In a one-class classification problem trained solely on benign data, FlowGuard reduces Attack Success Rates from >90% to <15% on unseen attacks, with <3% utility loss and up to a 6 times latency reduction. Our results demonstrate that monitoring cross-modal consistency offers an efficient and effective defense for multimodal reasoning.

cs.AI↗

arXiv2025

TRAP: Targeted Redirecting of Agentic Preferences

Autonomous agentic AI systems powered by vision-language models (VLMs) are rapidly advancing toward real-world deployment, yet their cross-modal reasoning capabilities introduce new attack surfaces for adversarial manipulation that exploit semantic reasoning across modalities. Existing adversarial attacks typically rely on visible pixel perturbations or require privileged model or environment access, making them impractical for stealthy, real-world exploitation. We introduce TRAP, a novel generative adversarial framework that manipulates the agent's decision-making using diffusion-based semantic injections into the vision-language embedding space. Our method combines negative prompt-based degradation with positive semantic optimization, guided by a Siamese semantic network and layout-aware spatial masking. Without requiring access to model internals, TRAP produces visually natural images yet induces consistent selection biases in agentic AI systems. We evaluate TRAP on the Microsoft Common Objects in Context (COCO) dataset, building multi-candidate decision scenarios. Across these scenarios, TRAP consistently induces decision-level preference redirection on leading models, including LLaVA-34B, Gemma3, GPT-4o, and Mistral-3.2, significantly outperforming existing baselines such as SPSA, Bandit, and standard diffusion approaches. These findings expose a critical, generalized vulnerability: autonomous agents can be consistently misled through visually subtle, semantically-guided cross-modal manipulations. Overall, our results show the need for defense strategies beyond pixel-level robustness to address semantic vulnerabilities in cross-modal decision-making. The code for TRAP is accessible on GitHub at https://github.com/uiuc-focal-lab/TRAP.

cs.AI↗