For others curious about AIs use in security, read up on the DARPA Cyber Grand Challenge. It started in 2016 at DefCon in Las Vegas and emulated a capture the flag game commonly played by human hackers at the event. This was long before generative AI and there were multiple teams leveraging AI to perform automated defense against attacks. AI has come a long way since that time but it was proven even in 2016 that AI could aid in finding vulnerabilities, creating patches for them and successfully launching attacks. You’d be surprised at what’s possible in security with AI. I was at that event and it was ridiculous how good these automated processes were even back in 2016. That required a lot of human training but with GenAI a lot of that goes away. As yourself this, if GenAI can build source code in a fraction of the time as a human and can access gigabytes of source code, why would you think it couldn’t automatically identify vulnerabilities and TTPs of attackers? It absolutely can. The bigger issue is that the attackers are also using AI so at some point it becomes a battle of who has more time or money. Though that’s really how it’s always been anyway. 😁
I've been hearing that #LLMs can identify anomalies and detect threats in every security vendor. However, in practice, LLMs primarily serve to contextualize and explain results. To date, yet to see LLMs performing root cause analysis, and I am skeptical that current transformer architectures are capable of achieving this. While hallucinations might be acceptable in marketing contexts, they are unacceptable in #security or monitoring applications. #AI has tremendous potential, but for mission-critical use cases, a complete overhaul of the existing architecture appears necessary. #ebc Thoughts?
Splunk had some time ago already the means to monitor metrics and look for anomalous activities and report them accordingly. It didn't take AI to perform such analyses. Even monitoring for example some of the thousands of SNMP metrics available on Linux servers can be very informative. That said, AI will open all sorts of amazing possibilities, for sure!
Generative Art Deco
1moThe optimistic view is that so far both humans and AI are failing equally and are therefore on par with each other at detecting threats or security vulnerabilities.