Concept detail
Prompt injection
6 tagged videos · 5 connection videos · degree 0
Publication timeline
- 2025-121 video
- 2026-085 videos
- 2026-102 videos
Source videos
Observed neighbors
external security boundary
motivates need for
1 source videos
Security Firewall for Agents — Ryan Dahl, DenoLethal Trifecta
Untrusted-content exposure is the injection vector; private-data access plus egress is what turns injection into exfiltration.
1 source videos
YOLO Mode, Safely: MicroVM Sandboxes for Any Agent — Rowan Christmas, DockerMCP/JSON-RPC tool integrations
Injection can enter through tool responses and protocol messages, not only through direct user prompts, expanding the attack surface to the framework layer.
1 source videos
AI Agents Gone Rogue? Build, Defend & Attack AI-Enabled Apps • Katie Paxton-Fear • YOW! 2025OpenClaw and MoltBook security failures
Obvious prompt injections, combined with exposed Supabase keys, caused the data leaks.
1 source videos
PyData Chicago Meetup: August 2026Thought poisoning
Extends the injection surface from user-visible text into the replayable reasoning layer, bypassing instruction-hierarchy defenses aimed at text inputs.
1 source videos
Why Frontier AI Labs Fight to Hide Chain of Thought — Ilia Shumailov & Alexander Panfilov