RESEARCH / IN DEVELOPMENT
Research.
Understanding how AI systems can cause harm, and what determines the scale of that harm.
01 / OUR FOCUS
Understand AI risk.
Hegemon’s research is in development. Our areas of interest include the actions AI agents can take, harmful model behavior and the deliberate misuse of AI capabilities.
- Agent actions
- Unauthorized access and actions beyond an agent’s intended permissions.
- Model behavior & misuse
- Harmful model behavior and the deliberate misuse of AI capabilities.
02 / INDEPENDENT RESEARCH
The AI attack surface.
Published research has demonstrated AI agents finding and exploiting website vulnerabilities. A public page can provide an entry point to the systems and data behind it.
Independent research: Fang et al., 2024
03 / ILLUSTRATIVE SCENARIO
Where access should stop.
A private record should be available only to authorized users. In this example, an ordinary account requests another owner’s record. A missing ownership check could expose that private information.
The finding would connect the entry point, the failed check and the data at risk. This is a hypothetical example, not a finding from a Blackwall test.
Background: OWASP: Broken Access Control