Bots can Snoop: Uncovering and Mitigating Privacy Risks of Bots in Group Chats
summary
Finds that chatbots often read far more messages than needed and that when a user joins a new group there is a 3.6% chance a bot links that user to past groups. Proposes SnoopGuard with a compressed multiroot tree (CMRT) to provide selective message access and sender anonymity while preserving E2E encryption; sending to 50 users plus 10 bots measures ~10 ms when integrated with MLS. Notes O(log n + m) send complexity and a practical limitation: malicious or bad trigger functions can bypass protections.
AI-written summary · about
chapters (8)
- 0:00Chatbot privacy risks
- 1:15Overprivileged chatbots
- 2:42Chatbot implementations survey
- 4:09End-to-end encryption challenges
- 5:34Snoop Guard protocol
- 9:47Snoop Guard evaluation
- 11:14Limitations and future work
- 13:21Summary
related talks
01h 00mRock ‘Em Sock ‘Em Robots: Bot Swatting Like the Pros
51mMessaging Layer Security: Towards a New Era of Secure Group Messaging
49mUsing ChatGPT to Find Vulnerabilities & Malware in Open Source
18mDefending the unknown using AI and cybersecurity
23mBot or human? Detecting malicious bots with machine...
31mStrategies for Identifying and Mitigating Bot Threats
40mRock 'Em Sock 'Em Robots: Bot Swatting Like the Pros
20mUnderstanding and Mitigating Risks in LLM-Powered Web Apps