The CISO's Dilemma: Why Banning AI Fails and How to Build a Governed Path

At the GovWare 2026 conference in Singapore, held on October 14, 2026, the conversation around enterprise security and artificial intelligence reached a critical inflection point. Chris Drake, the Founder and CEO of Armor, delivered a compelling Tech Talk Spotlights session titled “Once Again. AI Puts CISOs Between a Rock and a Hard Place.” His address cut through the industry noise, offering a pragmatic blueprint for security leaders grappling with the rapid, often unauthorized, integration of AI tools across their organizations. Drake argued that attempting to ban AI is a failing strategy; instead, organizations must establish a governed path that allows AI to complete useful business work automatically within policy.

The Cloud Transition vs. The AI Infiltration

To understand the unique challenge that AI presents to modern Chief Information Security Officers (CISOs), Drake contrasted the current AI wave with the historic transition to cloud computing. During the cloud migration, security teams maintained a degree of gatekeeping authority. CISOs could systematically assess cloud providers, evaluate security postures, and directly influence which applications and corporate data were permitted to leave the on-premises environment.

In stark contrast, AI does not wait for a formal procurement or security review. It arrives pre-packaged inside existing web browsers, everyday productivity tools, and directly on employee machines. This silent infiltration occurs without a separate, dedicated security decision. Depending on the permissions granted, an AI agent can read local files, transmit sensitive corporate data, and execute actions autonomously. Crucially, Drake pointed out that this risk bypasses traditional perimeter defenses; the data transfer can travel over an allowed connection even when the corporate firewall is functioning exactly as configured.

The Local VM Visibility Gap: Why EDR is Not Enough

One of the most technical and alarming aspects of Drake's presentation focused on the visibility gaps occurring at the endpoint level. For decades, organizations have relied on Endpoint Detection and Response (EDR) agents to monitor and secure employee laptops and workstations. However, the architecture of modern AI tools renders traditional EDR blind.

Drake highlighted specific documentation from leading technology providers to illustrate this gap:

• Anthropic documents reveal that host EDR solutions are unable to inspect activity occurring inside Cowork's local virtual machine (VM).
• Docker documents confirm that processes running inside its local sandbox microVMs remain completely invisible to the host operating system and its security agents.

Because of these local VM visibility gaps, EDR alone cannot provide the comprehensive oversight required to secure enterprise data. Drake emphasized that security leaders must look beyond the endpoint, utilizing platform telemetry and robust network controls to maintain visibility. When an employee clicks “Allow” on a local prompt, it does not mean the business has formally approved that data access or processing.

Establishing a Governance Framework for CISOs

The pressure on CISOs is immense. Drawing from his extensive conversations across the Asia-Pacific (APAC) region, Drake noted that security leaders frequently report to the very Chief Information Officers (CIOs) or Chief Technology Officers (CTOs) who are actively sponsoring and pushing for rapid AI rollouts. This reporting structure can place CISOs in a difficult position, balancing the business's desire for speed against the necessity of risk mitigation.

To navigate this pressure, Drake urged CISOs to take the lead and build trust by establishing a clear governance framework. This framework should not focus on approving every routine AI action, but rather on setting the high-level conditions for safe adoption. Drake outlined several essential components for this governance model:

• A written mandate that grants the CISO explicit authority to approve, restrict, or pause AI usage across the enterprise.
• A designated, named executive who takes formal ownership of each specific risk exception, along with a defined expiration date for that exception.
• A clear, structured escalation path directly to the organization's risk committee when conflicts between innovation and security cannot be resolved.

By establishing these parameters, CISOs can transition from being perceived as blockers to active enablers of secure innovation.

The Supply Chain and Fourth-Party Risk

The risk of AI adoption extends far beyond an organization's internal network, stretching deep into the vendor supply chain. When an enterprise partners with a third-party vendor, that vendor may utilize other AI providers to process the organization's data, introducing what Drake termed “fourth-party risk.”

To manage this exposure, organizations must demand absolute transparency and evidence from their suppliers. This includes knowing:

• Which specific AI providers are involved in processing the data.
• Exactly where the data is being processed and stored.
• What data retention policies are in place.
• Which contractual protections are actively applied to safeguard the information.

The scale of this visibility challenge is underscored by industry data. Protiviti's February 2026 AI Pulse Survey revealed that 47% of large organizations reported having incomplete visibility into their employees' use of AI tools. Drake noted that providing a useful, approved, and secure AI path is the most effective way to reduce the incentive for employees to bypass security controls and use shadow AI.

The Data Classification Imperative and Routing Control

At the core of any successful AI security strategy is robust data governance. Drake declared that data classification has transitioned from a “nice-to-have” operational practice to an absolute “must-have” requirement. “AI cannot safely use data you cannot govern,” Drake warned.

Organizations must have a precise understanding of:

• What their data contains.
• Who owns the data.
• How the data is permitted to be used.

Once this foundation of classification, ownership, and permitted use is established, organizations can implement automated rules to enforce policy.

Furthermore, Drake argued that organizations must own their routing policies. When data policies permit, the system can route queries to approved external models. Conversely, when dealing with highly sensitive or regulated data, the system must enforce private execution. The choice among these permitted options should be guided dynamically by task complexity, model capability, business needs, and cost. This approach allows AI to complete authorized work automatically, while generating detailed “receipts”—records of accessed data, policy decisions, models used, routes taken, actions executed, and final outcomes—allowing the organization to reconstruct events without requiring manual human approval at every single step.

Sovereign AI: Armor's Internal Solution Turned Enterprise Platform

Armor, a company that has been securing regulated industries since 2009, faced these exact challenges internally. Drake shared that the company was highly concerned about the security implications of sending sensitive customer data through public AI services, where controls and audit evidence could not guarantee compliance with their strict customer commitments.

To solve this, Armor built a governed AI platform for its own operations, which has now been commercialized as Sovereign AI. Drawing on 17 years of experience protecting over 1,700 organizations across 40+ countries, Armor designed Sovereign AI to provide a single control layer for every model, team, and dollar, operating entirely within the organization's own walls and under its own rules.

Drake outlined the core principles that guide the Sovereign AI platform:

• Know the data.
• Preserve model choice.
• Own routing policies.
• Control actions.
• Demand full auditability.

By providing a powerful, approved AI environment, companies can significantly reduce the appeal of shadow AI. Drake advocated for running powerful AI agents and their credentials within governed, isolated, and auditable environments with scoped access, rather than allowing them to run directly on employees' laptops.

Three Critical Questions for Modern CISOs

To help security leaders evaluate AI tools before granting them access to corporate environments, Drake urged CISOs to ask three fundamental questions:

• What can it touch?
• Where can it send?
• Can we control what it does?

By demanding clear answers, robust rules, proper authority, and auditable evidence, CISOs can confidently say “yes” to AI innovation without compromising the security or compliance of their organizations. His message to CISOs was clear: “Yes. With rules. With authority. With evidence.”

Armor demonstrated its Sovereign AI platform at Booth A22 during the GovWare event, which ran through October 15.