{"data":{"id":"08de380a-e2bc-4796-970f-36229b9594cd","title":"Depth-Dependent Safety Geometry in Vision-Language Models: Layer-Wise Localization and Bidirectional Refusal Control","summary":"Researchers present a lightweight inference-time framework for localizing and steering safety-sensitive representations in frozen vision-language models, without retraining. On Qwen3-VL-4B, steering along a learned safety direction in early vision layers gives bidirectional refusal control, with unsafe rejection spanning 6%-96%. The authors report that deeper layers stay separable along that direction yet respond little to steering, and that InternVL2.5-4B shows similar depth-dependent geometry.","solution":"N/A -- no mitigation discussed in source.","labels":["research","safety"],"sourceUrl":"https://arxiv.org/abs/2610.32182v1","publishedAt":"2026-10-23T05:28:49.000Z","cveId":null,"cweIds":null,"cvssScore":null,"cvssSeverity":null,"severity":"info","attackType":["jailbreak"],"issueType":"research","affectedPackages":null,"affectedPackageNames":null,"affectedPackageRefs":null,"affectedVendors":[],"affectedVendorsRaw":["Qwen3-VL-4B","InternVL2.5-4B"],"classifierModel":"claude-haiku-5-5","classifierPromptVersion":"v4","summaryPromptVersion":"v2","headline":null,"headlinePromptVersion":null,"cvssVector":null,"attackVector":null,"attackComplexity":null,"privilegesRequired":null,"userInteraction":null,"exploitMaturity":null,"epssScore":null,"epssCheckedAt":null,"kevDateAdded":null,"advisoryAliases":null,"affectedPackagesSource":null,"affectedPackagesCheckedAt":null,"patchAvailable":null,"disclosureDate":"2026-10-23T05:28:49.000Z","capecIds":null,"crossRefCount":0,"attackSophistication":"advanced","impactType":["safety"],"aiComponentTargeted":"model","llmSpecific":false,"classifierConfidence":0.93,"researchCategory":"preprint","atlasIds":["AML.T0054"]}}