Project summary

Ruhusa is an open-source authorization framework for AI Agents and multi- agent systems (https://github.com/Claire56/ruhusa). When a model makes a bad decision, follows malicious instructions or re-plans after denial , Ruhusa can stop the action from turning into a real world action.

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

1. To create a public benchmark for AI-agent authorization attacks. Example scenarios will include replanning after denial, escalating privileges for delegetated tasks, replaying actions, honouring authority revocations, replaying, reusing permissions across related tasks and tool substitution.

2. Run controlled experiments with model-driven agents to measure how often the agents attempt unauthorized actions and how often those attempts succeed.

3. Compare systems that rely mainly on model instructions with systems that enforce authorization independently at the point of tool execution.

4. I want an opportunity to , publish the benchmark, evaluation code, attack scenarios. In addition I want to publish, relevant results and documented failure modes so other researchers can reproduce and extend the work.

I’ll use Ruhusa as the authorization layer and will run repeated experiments across different prompts, task types, and agent behaviors.

How will this funding be used?

1. Support costs for Model/API evaluation runs

2. Maintain cloud execution infrastructure

3. Packaging the benchmark output and support for reproducing the various scenarios.

4. Pay for observability, tracing, storage, and evaluation tooling

5. Pay executions runs for Continuous Integration, Continuous Deployment and automated regression testing

6. Documentation and publishing infrastructure

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

I am the primary researcher and maintainer of Ruhusa.

I’m a PhD student in Artificial Intelligence studying authorization and security in multi-agent AI systems. I have professional experience building production agentic AI systems and other AI-based infrastructure.

Ruhusa has public repository on GitHub and is available as a package PyPI repository. I have already built the authorization framework, threat model, attack benchmarks, and automated security tests. The test cover delegation, replanning, revocation, replay, tool identity, confused-deputy risks, and authorization at execution-time.

This funding would support the experimental phase rather than starting a new project from scratch.

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

1. One possibility is that external authorization may work well against simple attacks but fails to provide protection in more complex agent workflows.

2. The benchmark scenarios are too synthetic and do not represent real-life agent failures.

3. There is also a usability tradeoff, an authorization system that's restrictive would make the agents unusable.

I’ll address these risks by testing both allowed and denied actions. I will measure false denials as well as successful attack prevention, using multiple attack group types, and publishing limitations alongside the results.

Even a negative result would be useful. Failure of a given authorization approaches would help identify where model-level or other system-level safeguards and fallbacks are still needed.

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

As of now i haven't raised any money yet.