AISECURITY SUMMIT26 SF

Classical security was built for human-timescale attacks and human-timescale defenses, but that foundation is being tested as AI enables attacks that operate at machine speed. In this research panel, leading academic security researchers cut through the industry noise to share what the data actually shows: where AI vendors' security practices and public narratives diverge from what's really happening, why systems-level thinking (not a pure-AI lens) is essential to this problem, and how rapidly improving open-weight models are simultaneously democratizing red teaming and lowering the bar for unsophisticated attackers. Grounded in original research rather than marketing claims, this conversation gives security leaders a clear-eyed view of where the frontier actually stands, and what "security by obscurity" gets wrong.

Open source security has hit an inflection point: AI models can now find and fix vulnerabilities at a scale no maintainer community can match by hand, and many organizations are turning to commercially patched distributions for support. Liran and Punit unpack where Project Glasswing has moved the needle most, what best practices are emerging, closing with a candid look ahead at the future of open source: can maintainers keep pace, or does secure open source become something you buy?

Software factories run on agents, and agents run on models. As teams choose between closed APIs and open-weight models they can inspect, fine-tune and host themselves, the security question shifts from the model to the pipeline around it. Faisal Tameesh (NVIDIA) and Jeff Wang (Cognition) discuss where open models earn trust, where agents need guardrails, and what leaders should decide now.

AI spend inside enterprise IT went from 1% to 10% in a year, and half of that is now security: one line in a much bigger reset. Software pricing is shifting from seats to usage, cost management is catching up with what these models actually cost to run, and open-weight models are closing the gap on the frontier. Add governments intervening directly on model access and chips, a regulatory regime still being written, and the oldest question in software moving at a new speed: vendor market or in-house, as LLMs speed up how fast anyone can ship. This panel brings together investors from first-check to public markets for an unfiltered read on what each of them is actually seeing.