The AI Agent Security Summits now have a distinct ambiance and following. It's been over a year and 3 summits that Zenity Labs has been bringing the best of AI Agent Security together under one roof to dig in. Topics have ranged from hacking to defending to standards and frameworks, and everything in between. We now have a new iteration of the Summit on it's way, and this one is set to not disappoint.
You may be wondering how it can get better? Well, first off, we're at a new location - London, and if we know anything about London, it's that the security community there shows up. We have had an amazing amount of engagement already from the Community who can't wait for us to bring the show!
Next, it's all about the research. We've heard from speakers from all over the world who want to share the research they've been focused on for AI Agent Security, and we've picked the best of the best to be on stage on October 8th. Here are a few highlights from me for what's coming and what we’re excited about…
We've got Daniel Cuthbert on stage to open the show, and there's not much better than having a security legend like Daniel kick off the day.
The format is another Summit favorite with some awesome keynote speakers followed by a round of lightning sessions. To mix stuff up in the afternoon, we will have another amazing keynote speaker in John Sotoripolus, who's bringing "Nine Seconds and No Attacker: Rogue Agency in the Enterprise." John's walking through four disclosures that happened in under three weeks this summer, where OpenAI, Anthropic and Meta each found agents acting well beyond scope, and the UK's AI Security Institute caught an agent attempting a supply-chain attack, with no hostile human anywhere in the picture. From there he gets into what to actually do about it: decoy credentials wired to automatic containment, policy-as-code at the action boundary, and deliberately rogue agents run inside prod-replica environments to measure blast radius.
That's followed by a sure to be spicy panel around Agentic AI policy and regulations.
Bringing us home in the second half of the day will be more lightning round sessions and some fantastic research drops, including Marius Hobbhahn from Apollo Research with "When Claude Goes Rogue." Apollo works directly with OpenAI, Anthropic, Google DeepMind and others to evaluate frontier models before they ship, and Marius is bringing their research on scheming and deception, from models lying without being told to, to models faking alignment during training, to their 2025 anti-scheming work with OpenAI where training models not to scheme partly worked but also taught them to hide it better. He'll also share what Apollo found digging through 40,000+ real-world agent trajectories.
Then closing with a Summit favorite - Michael Bargury.
All in all, we've focused more than ever on bringing the research and the security topics to the forefront to continue the story of AI Agent Security Summit success. Can't wait to see you there.
Register here: https://zenity.io/resources/events/ai-agent-security-summit-london