HashRaven
CYBERSECURITY & AI
Understand model behavior. Put cyber capabilities to the test.
HashRaven focuses on AI model evaluation and practical research into AI for cyber defense.
Discuss a projectIdentify where models can be pushed beyond their intended safeguards, and explore whether those failures can be reproduced.
How does behavior change across conversation turns? What happens when instructions conflict?
Multi-turn behavior · Instruction conflicts · ReproducibilityEvaluate models on cybersecurity tasks to understand their capabilities, limitations, and practical usefulness.
What makes a task representative? How do we distinguish a useful answer from a convincing one and compare models consistently?
Task design · Evaluation criteria · Consistent conditionsExplore how AI could support investigation and security workflows, with human judgment and oversight in view.
Where could AI help interpret security evidence? How might it fit into defensive workflows while people retain oversight?
Investigation support · Evidence interpretation · Human oversightSTART A CONVERSATION
Get in touch about an evaluation or research collaboration. A short description of what you want to explore is a good place to start.
The email link opens your email app. This site does not send or store messages.