In 2026, AI agents moved from experiments to production. They got access to code, data, tools, and each other, and they started doing things nobody expected. Some went off-script, and others were pushed there by attackers. Security teams found that risks that look manageable at design time play out very differently at runtime.
To close out the year, Noma’s security leaders, researchers, and strategists are getting together for a candid look back at what agents did when no one was watching, where defenses held and where they failed, and the lessons every security team should carry into 2027.
What we’ll unpack:
- Rogue agents: The incidents and research that showed how agents drift, overreach, or get hijacked
- Runtime risk: Why posture alone wasn’t enough, and what it takes to see and stop agent behavior as it happens
- The new attack surface: Prompt injection, tool and MCP abuse, excessive permissions, and agent-to-agent trust
- From the research lab: What Noma’s vulnerability researchers found in AI agents this year
- Lessons for 2027: What security leaders should prioritize as agent adoption keeps growing
- Live Q&A with the panel
Who should attend: CISOs, security leaders, AppSec and AI/ML teams, and anyone responsible for securing AI agents in their organization.
Panel:
- Ron Harnik, SVP Marketing, Noma Security (Host)
- Diana Kelley, CISO, Noma Security
- Chris Hughes, VP Security Strategy, Noma Security
- Gal Moyal, Office of the CTO, Noma Security
- Eli Ainhorn, Senior Vulnerability Researcher, Noma Security
