7ASecurity is proud to share the publication of our lightweight threat model for Super Tanks. Super Tanks is an open-source governance layer for autonomous AI agents, built around a local Python, Docker, SQLite, Ollama, and Telegram-oriented stack. The project presents a defense-in-depth control model that mediates agent actions before they reach tools, models, memory, code, smart-home integrations, or external systems.
The maintainer design goal for Super Tanks is clear and ambitious: prevent unsafe AI-agent actions before execution rather than trying to detect them after the fact. That mindset was visible throughout the engagement. Our review documented a credible defense-in-depth architecture with meaningful controls around gateway enforcement, identity, approval workflows, memory, code quarantine, integrity checks, operational modes, and auditability.
The public report is designed to help the Super Tanks maintainers and community reason about trust boundaries, attack surface, and future hardening priorities for agentic systems. This is exactly where threat modeling is most useful: before issues become real-world problems, while architectural decisions can still be strengthened deliberately..
"We built Super Tanks on a simple principle: unsafe agent actions should be stopped before they execute, not explained after the damage is done. An independent threat model isn't a stamp of approval - it's a map of where to look. We're grateful 7ASecurity gave us that map, and we're using it to guide the next round of hardening."
- William Park, Founder, KNDW Shelter Solutions AS
Threat Model Process
In June 2026, a team of 2 senior auditors from 7ASecurity carried out this whitebox lightweight threat model. The engagement dedicated 1.35 working days and used access to project documentation and source code, with limited source-code confirmation. This was not a penetration test, and no running validation was performed.
The scope was organized around the following areas:
- Super Tanks v3.2, commit c99a908
- Core security architecture
- Published threat model
- OWASP Agentic mapping
- STRIDE-based review of relevant assets, threat actors, trust boundaries, attack surface, countermeasures, and future hardening priorities
Threat Model Highlights
- Super Tanks implements a credible defense-in-depth architecture for AI-agent governance.
- The design shows clear attention to preventive controls, fail-closed behavior, containment, auditability, and safe composition.
- Positive controls observed in the review include gateway chokepoint enforcement, identity verification, DIQ role checks, per-agent allowlists, GO-Gate approval, ZEF re-scanning, memory RBAC, code quarantine, SAFE_MODE behavior, and audit-chain support.
- The documented architecture maps cleanly to STRIDE and the OWASP Agentic Top 10 categories, with threat boundaries reflected in implemented controls rather than only design prose.
- The project maintains substantial automated coverage, with 1398 tests reported as passing.
The Super Tanks team was helpful and responsive throughout the engagement, ensuring that 7ASecurity had the access and context needed to complete the review efficiently. This kind of collaboration is essential for productive security work, especially in fast-moving agentic AI systems where design intent, operational assumptions, and implementation details all matter.
Acknowledgements
Thank you to the individuals and groups that made this engagement possible:
- William Park and the Super Tanks team
- KNDW Shelter Solutions AS
- 7ASecurity: Abraham Aranguren and Noelia Aranguren
Read the report
You can read the public Super Tanks lightweight threat model HERE
If your team is building agentic software and wants practical security input before design decisions harden into production risk, talk to 7ASecurity.
