AI is a Trust Strategy.
Walking away from the Gartner conference, one message stood out: AI become part of how organizations operate, make decisions, serve customers and manage risk.
For the past several years, much of the conversation has centered on productivity. AI making employees perform tasks faster, automate repetitive work, analyze information, generate content or improve customer experiences. While those benefits still matter, the more consequential question is what happens when AI begins acting, not just assisting.
The shift toward agentic AI changes the equation. As systems move from generating recommendations to orchestrating workflows, interacting with other systems and executing actions, the traditional boundaries between technology, business processes, security, privacy and governance begin to blur. This message that was reinforced by the day’s keynote.
The next phase of AI is about agency. An assistant can help someone make a decision; an agent may participate in the decision, initiate another process, interact with systems or trigger an action. As such, the risk profile changes accordingly. The question is no longer simply whether an AI model is secure, but leaders must understand:
· What information can it access?
· What decisions can it make?
· Which systems can it interact with?
· What actions can it execute and within what boundaries?
· What happens when multiple agents interact?
· Who remains accountable for the outcome?
· Can we reconstruct what happened, including why an action was permitted?
These are enterprise risk and leadership considerations. Security must evolve with the operating model. Traditional cybersecurity has largely assumed that people initiate actions and technology executes them. Agentic environments increasingly challenge that assumption that makes identity, data governance and authorization foundational. Permissions need to be more granular. Monitoring must increasingly account for intent and behavior not just network activity. Additionally, logging must provide enough evidence to reconstruct not only what happened but how the system reached a decision and why it was allowed to act.
A particularly useful point from the keynote was that becoming an agentic organization is not about deploying more AI, it is about deciding where AI can act and under what level of control. A key takeaways described a spectrum from human-led and AI-assisted work through AI-enabled, AI-led and selectively autonomous operations. That is an important distinction since not every workflow should receive the same degree of autonomy. The level of delegation should reflect the nature of the decision, the potential impact and the organization’s ability to govern it.
The principle thar resonated most is that the greater the autonomy, the greater the need for accountability.Strong governance can enable faster adoption because leaders understand where AI can operate, where human oversight is required and which actions should remain off limits.
Jasleen Kaur Sindhu’ keynote, “Harnessing Agentic AI for Organizational Impact,” emphasized G.U.I.D.E. framework as a practical way to think about embedding that governance into the operating model:
· Govern enterprise context — establish the approved data, knowledge, policies and boundaries an agent can use.
· Unify decision orchestration — coordinate decisions and workflow routing across agents, people and systems.
· Invoke approved services — limit access to authorized tools, services and integrations.
· Deliver controlled actions — apply human approval or guardrail-based execution according to risk.
· Evidence every outcome — maintain traceable records that support review, accountability and continuous improvement.
The framework reinforces an essential point that AI governance cannot exist only in a policy document. It must be part of the architecture and the workflow itself. A practice through identity, permissions, data controls, monitoring, human checkpoints and auditable decision trails. A maturity progression shifts from scoped context and human-approved actions to adaptive routing, guardrail-based execution and continuously evaluated evidence as autonomy increases.
This also changes how organizations should approach AI initiatives. Rather than approving isolated agent use cases one at a time, leaders should start with enterprise workflows. Considerations around what work is being delegated, where humans retain decision authority, which agents and systems participate, and what controls must follow the workflow end to end. The keynote’s central recommendation is to design for the enterprise, not for disconnected AI experiments.
The traditional build-versus-buy discussion focused on features, cost, integration and capability. With agentic AI, organizations must also ask where context lives, who controls orchestration, where decision logic is encoded and how much control the organization retains. Vendors can accelerate capabilities, but the organization should retain ownership of its context, policies and decision logic. Hence, build versus buy is increasingly an intelligence-control decision, not simply a technology procurement decision.
The goal should not be to create AI that can do everything. It should be to create AI that clearly understands what it can do, what it cannot do and when a person must be involved. The competitive advantage will not come from adoption alone but instead from trusted adoption. Customers must trust how their information is used. Leaders must know where autonomous actions are occurring while employees must understand when AI is influencing decisions. Boards must understand the resulting enterprise risk while the organizations must be able to demonstrate that these systems are governed responsibly. Security leaders play more important role by not simply protecting AI from threats but helping the organization use AI in ways that are secure, responsible, explainable and worthy of trust.
Cybersecurity cannot sit outside the AI strategy, waiting to evaluate what the business builds. Security, privacy, technology, legal, risk and business leaders need to design the rules of engagement together. The keynote made clear that the challenge is not just enabling more capable AI but also establishing the controls, accountability and evidence that make greater autonomy sustainable.