Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs
Researchers demonstrated that message transcripts in collaborative LLM decoding can disclose private document content not present in the public view or final answer. The finding highlights information leakage risks in split on-device and cloud LLM architectures.
why this severityResearch paper demonstrating a leakage vulnerability with no known real-world exploitation.
- first report
- 30 Oct 2026
- last update
- 30 Oct 2026
Timeline · 1
Sources · 1
Research and catalogues
- 2026-10-30Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs with a Leakage LedgerarXiv · arxiv.org/abs/2610.38126
Titles and links belong to their publishers. The summary above is written by gcdRadar in its own words; data is CC BY-SA 4.0.
Classification
- lane
- Research
- AI role
- Attack on AI
- categories
- Data leak
- severity
- 1 · Research / proof of concept
- targets
- Not applicable
- owasp llm top 10
- LLM02:2025 Sensitive Information Disclosure
- mitre atlas
- no technique for these categories
- dates
- happened 30 Oct 2026 · first reported 30 Oct 2026
- origin
- News & research feeds
- record
- confidence 0.90 · created 2026-10-31 02:32Z · updated 2026-11-14 00:22Z
Lane, AI role, categories and severity come from the classifier; the OWASP and ATLAS rows are derived from the categories. How these are assigned
Copy a brief
This record as a short Markdown brief, for a ticket, a chat or a coding agent.
the brief as text
# Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs https://gcdradar.vercel.app/incidents/answer-only-disclosure-private-context-llm-assistance-over-w · AIR-2026-1683 · closed 2026-11-13 - Lane: Research - Severity: 1 (Research / proof of concept): Research paper demonstrating a leakage vulnerability with no known real-world exploitation. - Fix status: Fix status unknown - First reported: 2026-10-30 - AI role: Attack on AI ## Summary Researchers demonstrated that message transcripts in collaborative LLM decoding can disclose private document content not present in the public view or final answer. The finding highlights information leakage risks in split on-device and cloud LLM architectures. ## What to do No action recorded; read the sources below and the vendor's advisory. ## Latest updates - 2026-10-30: Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs with a Leakage Ledger (https://arxiv.org/abs/2610.38126) ## Sources - 2026-10-30 arXiv: Answer-Only Disclosure: Private-Context LLM Assistance over Web APIs with a Leakage Ledger https://arxiv.org/abs/2610.38126 Verify against the vendor's advisory before changing production. Data CC BY-SA 4.0, https://gcdradar.vercel.app.
Related
- liveClaude Code + SDKs Stolen Claude API keys turn Claude Code into a remote implant
Research · since 12 Nov 2026
AI - liveRailway Railway Breach Leads to Stolen AI Keys and $740,000 in Unauthorized Usage
Breaches & attacks · Real loss · since Oct 2026 (approx.)
AI - closedMicrosoft 365 Agents SDK CVE-2026-109485 Microsoft 365 Agents SDK Information Disclosure Vulnerability
Patch now · Disclosed · since 10 Nov 2026 · CVE-2026-109485
AI - liveRAG SNAPRAG: Cost-efficient knowledge extraction attack on RAG systems via adversarial perturbations
Research · since 10 Nov 2026
AI - liveSGLang SGLang RCE Exploit Breaches NEAR AI, Steals Data From 18 Agents
Breaches & attacks · Exploited · since 8 Nov 2026
AI - liveClaude Code + SDKs Claude Code Inadvertently Leaks Credentials to Public Repositories
Research · since 10 Nov 2026
AI