Davis: AI agents need verifiable computing, not trust

Davis's argument builds on a shift in how enterprises have historically managed risk. The old model worked because humans remained the ultimate decision-makers, even as organizations trusted cloud providers, software vendors, identity systems and administrators. Agentic AI breaks that assumption: autonomous systems can retrieve information, make decisions, invoke external tools, collaborate with other agents and execute actions on behalf of the enterprise. That means each individual component may be secure while the enterprise still lacks a complete record of how a final action occurred.

Davis points out that traditional security technologies and logs can show who was authorized to act and whether an unusual event occurred, but they do not necessarily capture the full chain of instructions, inputs, decisions and actions that led to an outcome. That gap becomes more acute as agents act across organizational and system boundaries, where one agent's decision may depend on information generated by another. Without a verifiable chain connecting those events, an organization may know the outcome without being able to establish how it happened.

The proposed path is incremental rather than a replacement. Identity management, endpoint protection, monitoring and policy enforcement remain indispensable, but they become part of a broader model in which independent verification adds a foundation for confidence. The practical test Davis offers is whether an organization can periodically reconstruct consequential agent actions using the evidence its systems produce. Whether that capability becomes standard practice will hinge on how many enterprises treat reconstruction as a deployment prerequisite rather than a post-incident exercise — and for security teams, compliance officers and business leaders, that choice is likely to shape how much autonomy they can safely grant.