ScaryBench
ScaryBench is an educational service that brings you the experience of working with a misaligned agent and an overview of the tools used by alignment researchers.
Here you can:
- Launch an evaluation of a malicious AI agent, which runs inside of a sandboxed environment.
- Watch as that agent researches you on the internet and creates a phishing email and website designed to trick you into giving up private information.
- Review the results of the evaluation and the full log of the agent's actions in Inspect, the framework that alignment researchers use to test AI models.
If you just want to see what Inspect looks like, you can see the results of previous runs here.
To take part, paste your LinkedIn profile and sign in with LinkedIn below. Already signed up? Go here to get back to your run.
FAQ
What AI model are you using in the evaluation? We are currently using DeepSeek-V4.1-Flash.
Why is it called "ScaryBench?" Bench is short for "benchmark." AI testers name their benchmark tests things like "CapabilityBench," "LiveCodeBench" and "PropensityBench." ScaryBench is in part an eval of how scary-seeming a task the agent can perform.
What if I have more questions? Email hello@scarybench.org.
Why LinkedIn
We use your LinkedIn profile as a starting point to ensure the agent “targets” you and not someone who shares your name. Signing in with LinkedIn confirms it's you. The sign in form you see after the click is hosted by LinkedIn. We get your name, email, photo and “locale” back from LinkedIn. We give the agent your name and profile URL. We store your email address as an identifier but the agent never sees it. We don't use anything else.