- Role
- Founder
- Team
- Solo, with partner firms
- For
- Enterprises and open-source maintainers
- When
- Sep 2026 to now
- Status
- Active · 10 reporter credits
In short.
StateCtrl is short for state control. Every AI agent moves through states as it works: reading input, calling tools, holding memory, taking action. Each one is a point an attacker can target. StateCtrl builds agents with control over every state, and tests live systems for private clients and open-source projects for the holes in between.
Founded in Sep 2026 and run solo, with partner firms for chatbot poisoning and GRC and SOC compliance. Every engagement starts with scope agreed with the owner; agent waves map and probe, every finding is reproduced locally and reported privately. So far: 10 reporter credits across 6 open-source projects.
Agents
- GLM
- Daybreak
- MiMo
- Customised harness
Disclosure
- GitHub security advisories
- CVSS
What this means for your business.
Fixes you can act on
Every finding comes with the exact attack path, a proof of concept and a suggested patch, so your developers know what to change and why.
Made for AI systems
Agents, chatbots and AI-built code fail in new ways: prompt injection, leaked data, tools with too much access. Those are tested alongside the classic holes.
Private until fixed
Findings stay private. Scope is agreed in writing, nothing outside it is touched, and every patch is retested before the report closes.
All projects.
Get in touch.
Have a question or a project in mind? Send me a message.