Speaker Floodlight: BSides Vancouver Island 2025

[Music] My primary goal is to help customers and individuals derisk their systems and stay cyber safe. We are living in the world of AI where we're just talking about generative AI, chat, GPT and copilot. It's simple three principles lease privilege, zero trust, verify explicitly, and assume breach. Basically, you should see who have access to your system. Currently, professionally, I'm the red team director for the enterprise red team at Electronic Arts. But I do a lot of security research with AI systems, which is really sort of my my big passion. You basically sometimes might not be aware that you hand over your entire system to a large language model that cannot be trusted implicitly. An agent can modify

its own environment and reconfigure its security controls can allow list its own capabilities like running arbitrary commands by itself. I create a tools and it have been released and I scan over the internet and there's turn out like 900 or more endpoint response our VXL scanner and there are nearly 4,000 IP inside the VXN terminal. Also the ter inside the VXLAN terminal are some public address. Oh actually everyone could hijack this public address then can abuse by attacker. AI can do amazing things but it can also hallucinate or produce conflicting results which is a huge risk in security. >> One of the main reasons that hallucinations happen is because large language models are working from their

working memory instead of from the information and context that you give it. give large language models the ability and capability to say I don't know and to reward them for that. >> Another example that I've been researching is a large botnet of microick routers. These routers are installed all over the world in homes and small businesses. They've formed what we call a large proxy network which can be used for all sorts of things. People with these routers might wonder why their internet is so slow. Well, maybe they're doing a DDoS attack on Ukraine, or maybe their internet is slow because their router is doing some kind of credential stuffing or mailing out other threats. And then when I was

working on a bug bounty for a gumb gambling platform, I discovered that the password reset token was vulnerable to padding oracle and I was able to use it for account takeover. A lot of times when developers are using CBC encryption, they assume wrongly that the resulting plain text can't be manipulated by an attacker using a padding oracle or simply just by modifying the cipher text. You can cause predictable outcomes in the resulting plain text which you can sometimes use to tamper with strings that are meant to be secure. And that's why when you're doing a padding oracle attack, you can reveal the plain text of anything except for the first block because that first block depends on the initialization

vector which is hidden on the server. >> Pretty much everything most cyber attacks use DNS. It's several stages of the attack. It's hard to keep track of how many domains an organization has, let alone how many subdomains an organization has.