Can open-source data improve how we estimate AI Security Incident likelihood?

Estimating incident likelihood (and therefore risk!) is an important process in cyber security. It helps us quantify the expected loss over a timeframe, assess whether this is acceptable, and inform how we manage the risk (and, importantly for the cyber security industry, help inform how much $$ to allocate to security or insurance!).

Last year, Tania Sadhani worked with Mileva Security Labs, an AI security startup, to investigate whether (and how) we can build on existing cyber risk modelling frameworks to address the new (and augmented) risks that AI components introduce to a system.

In this virtual talk she'll cover the current state and terminology used in cyber risk, a summary about our findings and things that have changed since that initial investigation!

Recommended Reading
https://aisecurityfundamentals.com/research

We'll end on an open-ended discussion on how
1. These findings apply to less narrowly defined risks like X-risks
2. The possible role of benchmarks in AI risk likelihood estimation.

This talk is virtual
Registration recommended but not required
Google Meet joining info
Video call link: https://meet.google.com/vbq-ogov-bys
Or dial: ‪(NZ) +64 9 886 4035‬ PIN: ‪717 937 517‬#