When AI Keeps Trying: What Leaders Must Control

Persistence is one of AI’s most valuable qualities. An automated agent can repeat a task, try another route, and continue when a person might stop. That same persistence becomes dangerous when the objective is broad, access is excessive, or nobody is watching.

The business risk is not an AI system with bad intentions. It is a fast system pursuing a goal through weak controls.

Automation multiplies mistakes

Traditional automation can already turn one error into thousands of bad transactions or defective products. An AI agent adds the ability to adapt between attempts.

That makes familiar weaknesses more important: forgotten accounts, unsupported systems, excessive permissions, and backups that have never been restored. The agent or attacker only needs one usable path.

Prove recovery instead of trusting a green light

A backup dashboard may show success while the actual recovery process remains untested. Leaders should request evidence from an isolated restoration exercise:

• What systems and data were restored?
• How long did the process take?
• What failed during the exercise?
• Does the recovery time match the business’s tolerance for downtime?

Three questions leaders should ask

First, what are we protecting? Ask for a business map, not just a software list. Identify the systems that run billing, payroll, sales, customer service, and operations, along with any outdated or over-permitted components.

Second, can we demonstrate recovery? Ask to see the latest exercise and the documented lessons.

Third, what authority are we giving AI, and who is watching it? Identify every agent, the information it can reach, the actions it can take, and the points where a human must approve the next step.

Assign accountability before autonomy

AI governance needs an owner. That may be an internal leader, an outside partner, or a combined team. The role connects experimentation with experienced business judgment, security, and documented decision-making.

Organizations do not need to avoid AI to stay safe. They need to use it intentionally, limit its authority, monitor its behavior, and prove they can recover when automation goes wrong.

An AITS AI readiness review maps the agents, critical systems, controls, and recovery gaps that leadership should understand before expanding automation.

Adapted from The Digital Dilemma newsletter.