Agents expose the limitations of trust
SiliconANGLE Renee Davis
Opinion — commentary, not a factual news event.
AI agents can now act across systems on their own. That makes trust alone too flimsy; companies need proof of what happened.
Based on reporting by SiliconANGLE, Renee Davis — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Enterprise computing has always leaned on trust. Cloud vendors, software makers, identity systems and admins were expected to do their jobs, and humans stayed in charge when something mattered. Agentic AI shifts that balance. These systems can fetch data, make decisions, call tools, work with other agents and take actions for the business without a person in the loop for every step.
That creates a harder problem than simple performance. An agent may pull from several systems, hand work to another agent, trigger an outside service and authorize a financial move. Each piece can be secure and still leave the company without a clean, independently checkable record of how the final action came together. Old-school security tools can show access, authorization and suspicious behavior. They do not, on their own, rebuild the full path from instruction to outcome.
The article’s core argument is blunt: asking which platform to trust is the wrong question when software is acting autonomously. The better question is what evidence the system produces, and whether someone outside the vendor stack can verify it. That means shifting from trusted computing to verifiable computing, where the point is not just control but reconstructability.
That doesn’t mean throwing out identity management, endpoint protection, monitoring or policy enforcement. It means putting them inside a wider model. Organizations need to decide which agent actions matter enough to audit closely, such as payment approvals, production code changes or access to regulated data. For those actions, they should set the allowed behavior in advance, define when human approval is required and capture enough evidence to connect the original instruction to the result.
The evidence has to travel across systems, too. If an agent uses a database, delegates to another agent and calls an external service, the audit trail can’t stop at one application and still be useful. The record should show the initiating user or system, the permissions in force, the tools used, the approvals received and the changes made. Cryptographic signatures, hashes, time-stamped attestations and other proofs can help preserve integrity and provenance. But only if the organization decided what to capture in the first place.
My take — AI-written commentary, not fact-checked reporting
This is the part of AI everyone keeps side-eyeing and then promptly ignoring: autonomy without receipts is just a faster way to lose the plot. Agents are useful, sure, but the minute they can move money, code or regulated data, “trust us” becomes a joke with an audit trail problem. The grown-up move is boring and necessary: prove it, or don’t let it act.
Read more about this at: SiliconANGLE
