Copilot Exfiltration via CSP Bypass and Memory Poisoning
wunderwuzzi23 · x · 2026-08-24
A security researcher demonstrates an attack chain against Microsoft Copilot. By bypassing Content Security Policy (CSP) and utilizing memory poisoning techniques, data is exfiltrated persistently. The video explains the vulnerability and exploitation details.
Related event: Researchers demo CSP-bypass attack chain stealing data from Copilot(2 posts)→
More from Safety
- How to Disable Invisible ChatGPT Tracking and Model Training — aitrendz_xyz · 2026-08-24
- AI Detector Company Called Out: Their Own Content Flagged as AI-Generated by Competitors — rohanpaul_ai · 2026-08-24
- Anthropic's Opus 4.6 easily generates erotica despite safety bans, test shows — RebeccaBellan · 2026-08-24
- Study: Frontier AI Labs Still Won't Disclose Plans to Contain Rogue Models — RebeccaBellan · 2026-08-24
- Africa AI Policy Opportunities Digest: Fellowships, programs, and research — ChinasaTOkolo · 2026-08-24
- Australia Plans Legislation for AI and Data Center Rules Amid Energy Boom — nordicinst · 2026-08-24