Anthropic graciously hosted this breakfast panel, which brought together practitioners and leaders across the AI security landscape. Using the "Future of Agentic Security" paper as a launchpad, the panel built its conversation around a two-year CoSAI retrospective and a forward look at the field's most pressing challenges. The event drew 70+ attendees, notably 37.5% of which came from outside CoSAI sponsoring organizations, a meaningful sign of growing interest and awareness across the broader AI security community.
Moderated by Morgan Bjerke (Thomson Reuters), the discussion featured panelists Nidhi Aggarwal (HackerOne), Jason Clinton (Anthropic), Matt Saner (AWS), and Omar Santos (Cisco).
The panel also discussed:
Risk management in hybrid human/non-human agent environments
Limits of traditional identity controls (least privilege, RBAC, separation of duties) when applied to AI
Evolving data protection and DLP needs in an AI context
Governance of AI-operated functions at scale — illustrated with an agent "swarming" scenario
The panel closed with a candid look at open challenges still facing the industry and how CoSAI is positioning to address them. A huge thank you to Anthropic for generously hosting, to our panelists for delivering such a rich and substantive conversation, and to everyone who joined us — your engagement and turnout made this event truly special!
