The AI Security Podcast

This week we released the interim report and subset of the AI incident database for the AI Security Likelihood research Tania has been working on! We chat through the motivations for the project, our initial insights and takes, and where we hope to take it next. Now we want your input! Check it out at aisecurityfundamentals.com and let us know if you have insights on the risk modelling, AI incidents, or access to data/collaborations that could make this project even better.
(If you don't want to hear us talk about our Christmas party skip to around 8 minutes in).

Big thanks to Foresight for funding this research, and our collaborators at the Australian National University and University of New South Wales.

We mention this book, How to Measure Anything in Cyber Security and recommend giving it a read: https://www.oreilly.com/library/view/how-to-measure/9781119892304/cover.xhtml

Today we chat to my good friend Annie-Mei about GRC (governance, risk and compliance). She is a senior GRC consultant with a background in journalism! We discuss her career journey and an overview of what on earth GRC is.

This time we sit down with Howjer Gu, Founder of Planoverse, a start-up in our stream within the UNSW 10x accelerator! Fun fact - Howjer and I used to work together at Deloitte back in 2016 and couldn't believe it when we ran into each other during the application process. Howjer has a background in data analytics and recently founded Planoverse, which is like a digital twin for grocery stores. Learn about his start-up journey and the importance of data.

Planoverse: https://www.planoverse.com.au/

2024 has seen a lot of change for Mileva.. and we often get people asking about what we're doing and we got here, so this time Tania interviews Harriet about starting Mileva! We know lots of our audience are curious about starting a business and the different trade-offs that entails (quitting your full time job, looking for funding, services vs products, when is the right time) so we address some of these questions. We also talk about our experience in the UNSW 10x accelerator (hint: I'll be interviewing some of these founders in upcoming episodes!)
Oh also Harriet somehow ended up in Vogue!

In this episode Tania and I chat through how we learnt about AI security and adversarial machine learning, and how you can too! 

Some resources we mention:
- Crash Course AI https://thecrashcourse.com/topic/ai/
- 3 Blue 1 Brown https://www.youtube.com/c/3blue1brown
- Kaggle https://www.kaggle.com/learn/intro-to-machine-learning
- Hands-on ML textbook (more for 'classic' machine learning - still really important to understand 'modern' models) https://www.booktopia.com.au/hands-on-machine-learning-with-scikit-learn-keras-and-tensorflow-3e-aurelien-geron/book/9781098125974.html?gad_source=1&gclid=CjwKCAiA9dGqBhAqEiwAmRpTC3W7SbJJOrI8fEi-19OxvyNLElA8Nk7iKH6nE4KDm1iLlySKgb114hoCcG8QAvD_BwE
- The paper that started it all (or at least a lot of it) https://arxiv.org/abs/1312.6199
- Adversarial Robustness Toolbox https://adversarial-robustness-toolbox.org/
- Cleverhans https://cleverhans-lab.github.io/
- Darknet diaries (please someone introduce me to Jack) https://darknetdiaries.com/
- Rob Miles https://www.youtube.com/@RobertMilesAI
- Import AI https://jack-clark.net/
- AI Daily Brief https://www.youtube.com/channel/UCKelCK4ZaO6HeEI1KQjqzWA
- Less Wrong https://www.lesswrong.com/
- Unsupervised learning https://danielmiessler.com/
- OWASP ML Top 10: https://owasp.org/www-project-machine-learning-security-top-10/
- ML Safety https://course.mlsafety.org/about
- Anatomy of an AI system: https://anatomyof.ai/

People kept asking us about the recent AI laws in California and why some were passed but one was vetoed, so we decided we'd finally better do a podcast episode on it!
Disclaimers: we're not lawyers (this will be immediately obvious) but I hope this is enlightening to other AI/tech folk. We certainly loved researching this one, and we encourage everyone who's interested to join the discussion!

In this episode we shout out Civ.ai, a non-profit we met at Berkeley when we were in the US. They create demos to show policymakers just how impactful and risky AI can be, so please do get in touch with them if you're interested! https://civai.org/

In this bonus episode we're still with Miranda and we're talking about some of the trends we're seeing around LLMs (language models) being used by adversaries to augment attacks. This could be another episode, but we have a bite sized discussion for you here!

This time we are joined by another vital member of the Mileva team, Miranda R! Unlike Tania and myself, who are data scientists who got into security, Miranda actually started as a cyber security analyst and made her way into AI. She chats about her career journey and insights on the cyber/AI intersection.

When we were in the US we rode a Waymo.. and now we chat about it here! We definitely want to chat about job displacement in another episode, but for now - here is our excuse to tax deduct the Waymo ride!

We're finally back after a bit of a break (travel is tiring). Over the last couple of weeks we attended and spoke at BSides Singapore and BSides Canberra! In this episode we cover the research discussed in our talks and the highlights of BSides in general (you know we're massive BSides fans).