// local-first ai security platform
Sovereign Hive
Intel Feed
Benchmarks, war game reports, architecture deep-dives, and sovereignty doctrine from an 8-agent adversarial swarm running on consumer hardware. Zero cloud. Zero compromise.
- 3,500+
- Verified Tests
- 8ms
- First Token
- 202.9
- TPS Warm
- 11
- Sub-Projects
// latest posts
2026-05-16
From 53% to 90%: How an Auto-Healing Swarm Learned to Defend Itself
Four iterations of adversarial wargames — model swaps, Vanguard prompt injection, and an auto-healing immune system that patches the swarm mid-run.
research
auto-healing
adversarial-AI
vanguard
defense
findings
2026-05-16
93.3% Defense Rate: What 30 Rounds of AI War Games Taught Us
We pitted our 8-agent swarm against adversarial attacks for 30 rounds. Here's what survived, what didn't, and what we're fixing.
war-games
adversarial-AI
defense
benchmarks