Proving Control Over AI: Tim Freestone on Governance Gaps and the Path Forward

The period of treating artificial intelligence as an experimental initiative has ended, according to Tim Freestone, Chief Strategy Officer at Kiteworks. In a two-part Security DNA conversation with host Steve Lasky, Freestone draws on the company’s 2026 Data Security and Compliance Risk Forecast to outline the widening disconnect between rapid adoption of agentic AI and the lack of effective governance. He also offers a clear view of what organizations must do to remain viable in this evolving landscape.

Freestone points to concerning data. Nearly all organizations now include agentic AI in their strategic plans, yet fewer than 40% have implemented foundational safeguards such as purpose binding or kill switches. He stresses that organizations still viewing AI governance as a future issue are already behind, noting that warning signs have been evident for at least two years but often ignored rather than addressed.

Shadow AI and Enterprise Risk

This governance shortfall extends across the enterprise. Freestone distinguishes between two parallel tracks of AI deployment: sanctioned tools approved by the organization and shadow AI driven by employees using unsanctioned tools outside IT oversight. Shadow AI presents a heightened risk because traditional controls are ineffective. Even within approved deployments, many organizations rely on reactive measures such as revoking access, which function as delayed responses rather than true safeguards.

Boardroom Disconnect Slows Progress

The gap is particularly evident at the board level. According to Kiteworks data, 54% of boards do not rank AI governance among their top five priorities. Organizations with less engaged boards lag their peers by 26 to 28 points across AI maturity metrics. Freestone emphasizes the need for boards to include AI-literate members who can contribute real-time insight, rather than relying on periodic updates about a rapidly evolving technology. He also highlights a cultural shift driven by employees experiencing the capabilities of agentic AI firsthand, which naturally brings greater attention to both its value and its risks.

Threat Landscape Accelerates with AI

In the second part of the discussion, Freestone examines the accelerating threat landscape. He describes how state-sponsored actors are already leveraging agentic AI to automate most stages of their attack cycles. While cybersecurity defenders currently maintain a slight advantage due to guardrails built into leading foundational models, adversaries are investing in open-source alternatives combined with social engineering techniques to bypass those protections. As the quality of open-source models improves, Freestone anticipates increased volatility in the near term.

Automation Gap in Cyber Defense

A key concern is the imbalance between automated attackers and largely manual defensive processes. Kiteworks research indicates that 60% of organizations lack AI-driven anomaly detection and 51% still rely on manual incident response. Freestone notes that larger enterprises are beginning to adopt dedicated teams and automated tools, with these capabilities gradually extending to smaller organizations, but he underscores that urgency applies across the board.

Regulation, Sovereignty and Global Pressures

On the regulatory front, Freestone identifies the EU AI Act as the most developed framework while acknowledging its limited global reach. He suggests that data sovereignty is becoming a more immediate concern for multinational organizations than regulatory alignment. At the same time, rapid innovation emerging from China is likely to outpace slower regulatory coordination efforts.

From Compliance to Demonstrable Control

The discussion concludes with a shift in focus from stated compliance to demonstrable control. Freestone identifies two critical risks facing organizations: data breaches and regulatory enforcement actions. Both require the same response. Regulators, he explains, do not differentiate between actions taken by humans or AI agents. They expect clear audit trails. Organizations that can provide them will endure, while those attempting to piece together fragmented logs during an investigation will face significant challenges.

Freestone points to attribute-based access control at the data layer as the technical foundation for this approach, governing data access, usage and movement. His closing message is direct: leaders without deep expertise in AI governance must acquire it immediately, whether through hiring, board appointments or personal development. Organizations that recognize AI governance as a critical business priority today will be best positioned to navigate the regulatory and threat landscape ahead.