Project summary

Super Tanks is an open-source (Apache 2.0) runtime that enforces safety limits on autonomous AI agents at runtime — least-privilege action-gating, human-approval gates, interruptibility, and tamper-evident logging — plus an open red-team/eval harness that tests whether an agent escapes its constraints, mapped to the OWASP Top 10 for Agentic Applications.

What are this project's goals? How will you achieve them?

Close the gap between agents-everywhere and agents-we-can-constrain. Most safety work today is advisory; few open tools enforce limits at runtime or test whether an agent stays within them. We will ship (1) an open safety-control enforcement library and (2) an open agentic red-team/eval harness with attack scenarios mapped to the OWASP Agentic Top 10, plus reproducible results — all Apache 2.0.

How will this funding be used?

USD 25,000: open red-team/eval harness mapped to OWASP Agentic Top 10 — 10k; safety-control enforcement library (action-gating, interruptibility, oversight) — 9k; public Apache 2.0 release, docs and reproducible eval results — 6k. The 10k minimum funds the eval harness alone.

Who is on your team? What's your track record on similar projects?

William Louis Park — solo founder, KNDW Shelter Solutions AS (Norway); around 12 years of EPC/industrial project management. A working prototype already runs (1,331 passing tests) inside KNDW own multi-agent stack — policy gating, human-approval gates and tamper-evident logging in daily use. Honest status: private prototype, pre-release, not yet externally adopted.

What are the most likely causes and outcomes if this project fails?

Enforcement too coarse could cripple agent usefulness (mitigation: configurable least-privilege policies tested against the harness). Solo-founder bandwidth (mitigation: scope bounded to the eval harness plus control library, not a full platform). Standards drift as the OWASP list evolves (mitigation: modular, version-pinned mappings).

How much money have you raised in the last 12 months, and from where?

Innovation Norway grant 2026/130689 (compliance-by-design AI safety architecture). An NLnet NGI Zero application was submitted on 29 May 2026 (ref 2026-06-36d, pending) for the complementary compliance-runtime scope.