AGENTS OF S.E.A.L.E.D: AI Agentic Cybersecurity Framework
summary
Modular on‑prem agentic framework runs as ~12 Docker containers and orchestrates models across 6GB–24GB GPUs, using Gemma 270M intent model to pick services and swap to larger models like GPT‑OSS 20B CVE mapping or Gemma 3N for analysis; guardrails pipeline aggregates IBM Granite/Llama Guard/Gemma Shield (Q8 quantized) votes to block harmful prompts. Demo shows SQL‑injection detection, image→policy parsing that generates compliant passwords, graph storage via NetworkX + FAISS embeddings, TailScale for decentralized inference, and an imminent open‑source release.
AI-written summary · about
chapters (10)
- 0:00Introduction and backstory
- 3:33The problem landscape
- 5:47S.E.A.L.E.D framework design
- 10:50Encryption and policy management
- 11:57Analysis and threat modeling
- 12:54Elimination and defense
- 13:55Orchestration pathway
- 17:05Platform implementation demo
- 23:12Functional use cases
- 32:21Conclusion and future work
related talks
01h 42mBuilding Responsible AI Agents in Decentralized Open-Source Environments
35mGenAI Application Security: Not Just Prompt Injection
05h 56mFrom the main stage to the terminal
01h 07mEmpowering Security Teams with Generative AI
19mKeynote: Towards Building Safe and Secure Agentic AI
01h 07mSecure distributed multi-agent systems
30mNavigating the New AI Security Landscape
39mSecuring AI at Scale: Practical Defenses against Prompt...