TVDB-2612CRITICAL
Cross-agent memory poisoning via shared vector store
A single poisoned embedding steers every downstream agent that retrieves it. No direct access to any agent is required.
✓ DETECTION SHIPPED TO GUARD6H AFTER DISCOVERY
WHY US
Research-backed. Publicly tested. Sovereign.
01
AI Threat Vector Database
Zero-day agentic threats, researched in-house and shipped as detections.
02
LLM Security Leaderboard
6 public editions since 2025, with 14 frontier models scored on our own attack dataset.
03
Sovereign
Supports air-gapped deployments, private VMs and SaaS.
RESEARCH
Powered by our own zero-day agentic threat research.
TVDB is our threat-intelligence database of malicious prompts and attack techniques, collected as they appear in the wild and matched by semantic similarity. New entries ship as detections before they become incidents.
Explore TVDB →