Izzy Barrass

About Me

Currently: Founder of Aethra Labs - we provide safety evaluations, red teaming and agent runtime monitoring services to AI labs and companies deploying AI products.

We've run pre-deployment red teaming engagements with almost all of the frontier AI labs. Our safety evals are drawn from leading benchmarks developed by our team whilst at the Center for AI Safety.

Our runtime monitoring approach leverages eBPF and system level observability techniques, and applies this to the problem of tamper-proof agent runtime monitoring - we're the only runtime monitoring approach that can monitor agents without access to the source code, and are zero-code deployment.

Previously:

I studied at the University of Oxford and University of Melbourne.

• Founder: co-founded a red teaming company with Pliny the Liberator. I built out and hired the early team, and ran the first red teaming engagements with AI labs and various companies.
• Research TPM: in AI safety research, where I led the development of safety evals datasets and helped orchestrate/support research teams to meet tight deadlines. My favorite research project was developing the WMDP benchmark from scratch (CBRN eval) - where I managed 30+ expert researchers to develop a dataset of 4,000 MCQs.
• Strategy & investing: in a past life, I was a strategy consultant at McKinsey and growth investor at a $6B fund.

Find Me Online

• I don't like Twitter, so I don't use it
• LinkedIn

Email me: izzy (at) aethra-labs (dot) com