AI versus Marshy - A Crash-Course on AI Security

I’m Marshy, the guy behind this newsletter that keeps tabs on AI trends and breaks them down for you. I’ve been in the tech industry for a while, helping clients understand and use tech in their businesses. This article is a summary of a conversation I had with HarrietHacks, a cybersecurity expert who’s been creating content on AI security. That’s what happened when my cybersecurity friend mentioned HarrietHacks . Harriet started a YT channel on AI security last year and has released a steady stream of content since. A pay phone. I love hearing about this kind of stuff as it’s early days for Harriet’s content but as with any new thing it just takes time and consistency. The ingredients are there: ✅ Woman in tech ✅ Booming area (cybersecurity) ✅ Booming trend (AI) I reached out to Harriet and explained what I do with this newsletter and asked what are the main things people should considering when using AI in regards to cybersecurity. Harriet kindly shared four (which I’ve edited for brevity): The term AI is still foggy and often used inappropriately There have been a mixed bag of terms over 70 years for what AI is and is usually considered “cutting edge”. The other trap is sensationalisation about it taking over. ISO (the International Organisation of Standardisation) gave it a much simpler definition last year - something that can do what a human might have done, but using a model. So think about AI as a goal, rather than a technology. AI security is important but no one is talking about it It deals with the technical and governance practices that aim to secure AI systems from disruption, deception, and disclosure of information. There is a lot of talk about safety (what if its misused?) and not enough about the threat of bad actors/adversaries. AI security is different to cyber security Being cyber secure isn’t the same as AI secure and there are dozens and dozens of threats if an AI system is dealing with access to information, input, video, transactions etc. Most organisations aren’t aware of what’s necessary to protect themselves. The AI security threat is real Attacks are rising and the numbers are a bit blurry because organisations blend AI safety and security together. The threat is probably higher than reported (and it’s already high). Thanks Harriet! I think what blows my mind is the human element. When it comes to security the human is always the weakest link - so having ever-improving capabilities for deceiving humans is terrifying. Just over the weekend a Hong Kong finance worker paid $25m to fraudsters who imitated the company CFO using a deepfake . eek. AI as a thinking companion Via One Useful Thing . Heuristics, frameworks, rules of thumb - I love them and have a number of recipes saved in my notes folders. Originally appeared in newsletter : AI versus Marshy #33: security, thinking, and fakes