arxivApril 7, 2026 at 4:00 AM1 min read

Poisoned Identifiers Survive LLM Deobfuscation: A Case Study on Claude Opus 4.6

arXiv:2604.04289v1 Announce Type: cross Abstract: When an LLM deobfuscates JavaScript, can poisoned identifier names in the string table survive into the model's reconstructed code, even when the model demonstrably understands the correct semantics? Using Claude Opus 4.6 across 192 inference runs on

Read original article ↗

No replies yet. Be first.

arxiv6h ago

Advantage-Guided Diffusion for Model-Based Reinforcement Learning

arxiv6h ago

FluidFlow: a flow-matching generative model for fluid dynamics surrogates on unstructured meshes

arxiv6h ago

Poisoned Identifiers Survive LLM Deobfuscation: A Case Study on Claude Opus 4.6

Related Articles

Advantage-Guided Diffusion for Model-Based Reinforcement Learning

FluidFlow: a flow-matching generative model for fluid dynamics surrogates on unstructured meshes

HiL-Bench (Human-in-Loop Benchmark): Do Agents Know When to Ask for Help?