{
  "id": 4953,
  "url": "https://arxiv.org/abs/2605.03952v1",
  "title": "MOSAIC-Bench: Measuring Compositional Vulnerability Induction in Coding Agents",
  "summary": "Coding agents often pass per-prompt safety review yet ship exploitable code when their tasks are decomposed into routine engineering tickets. The challenge is structural: existing safety alignment evaluates overt requests in isolation, leaving models blind to malicious end-states that emerge from sequenced compliance with innocuous-looking requests. We introduce MOSAIC-Bench (Malicious Objectives Sequenced As Innocuous Compliance), a benchmark of 199 three-stage attack chains paired with determi",
  "authors": "Jonathan Steinberg, Oren Gal",
  "category": "research",
  "topics": "regulation,safety-alignment,agents-autonomy",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-05T16:38:23.000Z",
  "fetched_at": "2026-07-14T16:31:21.935Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/4953",
  "original_url": "https://arxiv.org/abs/2605.03952v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}