When AI Starts Acting: Inside Cybersecurity Implications of AI 2026
Twenty security leaders and practitioners examined how to govern AI agents, uncover hidden AI use and strengthen defenses without losing human accountability.
Approving an AI application is one decision. Deciding what it can do on the organization’s behalf is another.
Who is accountable for its actions? Which systems can it reach? How quickly can defenders recognize behavior that has moved beyond its intended purpose?
The Cybersecurity Implications of AI event, held August 20, 2026, brought CISOs, security architects, AI leaders and practitioners together to examine those questions through practical frameworks and operational lessons.
Across 12 sessions, the program connected two challenges that deserve to be considered together: securing AI systems and using AI to strengthen security.
Making Autonomous Agents Accountable
An agent’s ability to complete a task does not establish whether it can be trusted to complete it safely.
- Jessica Marie of Proofpoint examined how enterprises can manage AI agent risk through runtime integrity, intent alignment and a maturity model for advancing agent security.
- Sunil Arora of Humana presented frameworks for securing autonomous AI agents, using CIAA and MICE to address accountability, defense in depth, runtime controls and governance.
The useful test is not simply whether an agent performs well in a demonstration. It is whether the organization can define its boundaries, observe its behavior and establish who is accountable when something unexpected happens.
Finding AI Before It Becomes a Blind Spot
Before leaders can govern AI, they need a reliable view of where it is being used.
- Kevin Kiley of Airia explored how to bring expanding AI environments under control through discovery, security, governance and optimization, with attention to model risk, compliance and costs.
- Ben Shelston of Netskope examined how to uncover AI usage across the enterprise and apply consistent controls across applications, endpoints, cloud systems and agentic interactions.
Discovery and control need to work together. An inventory becomes useful when it helps the team decide what is acceptable, what needs tighter controls and what should not be running.
Giving Agents an Identity and Limits
A trusted identity is a starting point, not permission to act without boundaries.
- Alex Laurie, Brian Yoon and Manish Kalia of Ping Identity discussed how to establish trust in AI agents through identity, access controls and governance.
- Saurabh Srivastava of Hilton examined how to secure enterprise identity at machine speed, focusing on reducing defensive latency and governing non-human identities against agentic attacks.
Together, these sessions connect two requirements: knowing which agent is acting and being able to respond quickly enough when its access becomes a risk. Identity controls need to support both.
Defending at the Speed AI Enables
Securing AI and using AI for defense are different jobs. This part of the program examined how those responsibilities intersect.
- Charles Iheagwara, Payam Hojjat, Sanjit Ganguli and Vianney Chimezie discussed how to secure agentic AI while strengthening cyber defenses through governance, AI asset visibility, machine identity controls, behavioral detection and automated defense.
- Jesus Toledano of Google Cloud Security explored how agentic security operations could change the economics of zero-day breaches, including anticipating threats, validating controls in digital twins and deploying defenses at machine speed.
- Alon Schindel and Raaz Herzberg of Wiz examined how to address AI-accelerated vulnerability exploitation by reducing exposure, testing critical code and using AI agents to accelerate detection and remediation.
The shared challenge is not speed alone. It is making faster decisions with a clear understanding of the exposure being reduced and the controls governing the response.
Securing the Agentic Development Life Cycle
Security testing has to follow the development process as it changes.
- Said Syed of Snyk for Government examined how to secure the agentic development life cycle through continuous testing, agent governance and autonomous red teaming at machine speed.
This shifts the emphasis from a single approval checkpoint to continuing assurance. The question becomes how to keep testing and governing an agent as development and deployment evolve.
Keeping Governance Connected to Real-World Consequences
The appropriate level of autonomy depends on the environment in which AI operates.
- Sébastien Christophe Lacroix of The Coca-Cola Company examined how AI air gaps can connect cybersecurity and compliance across model supply chains, data sovereignty requirements and global restrictions.
- Ali Baccouche of Texas Instruments, Claudio Sangaletti of PERSÁN and Renata Araújo of Braskem discussed where AI can help operational technology security and where oversight remains essential, covering practical use cases, governance and human involvement.
These discussions put boundaries back into the conversation. Security teams need to consider where AI is deployed, which requirements apply and when human judgment must remain part of the decision.
Cybersecurity Implications of AI 2026, by the Numbers
- Six hours of focused cybersecurity instruction
- 12 sessions covering AI governance, visibility, identity, defensive operations, development and OT security
- 20 featured speakers across individual presentations and multi-expert discussions
- 10- to 45-minute sessions, averaging 30 minutes
Taken together, the program offers a connected view of AI security: know where AI is operating, limit what it can do, verify its behavior and preserve accountability as the pace of work accelerates.
Get Every Session and Keep Learning
These sessions are just the beginning. Security Insights by CyberEd brings global ISMG event replays, expert-led masterclasses and CPE-eligible sessions into one on-demand library.
The scale speaks for itself: 400 events annually, 75,000+ attendees, 500 expert speakers and more than 1,000 event sessions.
And there is no travel itinerary to build around it. For $495 a year, you can access the library on your schedule, without flights, hotels or days away from your team.
Stop choosing which event to attend. Get them all.