AI agents are now integrated into many core business functions globally. Soon, they may schedule our lives, make critical decisions, and negotiate deals on our behalf. While this is exciting, it raises a crucial question: who oversees these agents?

Login to Continue Learning

Over half (51%) of companies have already deployed AI agents, with Salesforce CEO Marc Benioff aiming for one billion by year-end. Despite their growing influence, verification testing remains absent. These agents are entrusted with critical responsibilities in sensitive sectors like banking and healthcare without proper oversight.

For efficient and accurate goal-oriented actions, AI agents need clear programming, high-quality training, and real-time insights. However, not all agents will be created equal. Some may receive more advanced data and training, leading to an imbalance between bespoke, well-trained agents and mass-produced ones. This could pose a systemic risk where more advanced agents manipulate less advanced ones over time.

For example, an AI agent might misdiagnose a critical condition in a child due to its training primarily on adult patient data. Or an AI agent chatbot could escalate a harmless complaint because it misinterprets sarcasm as aggression, slowly losing customers and revenue due to misinterpretation. According to industry research, 80% of firms have disclosed that their AI agents have made “rogue” decisions.

Unlike traditional software, AI agents operate in evolving, complex settings. Their adaptability makes them powerful but also more prone to unexpected and potentially catastrophic failures. For instance, an agent might misdiagnose a child’s condition or escalate a customer complaint due to data biases.

In a unique start to his ruling against President Donald Trump’s administration on September 30, U.S. District Court Judge William Young included a scanned handwritten note sent to his office. The note read: “Trump has pardons and tanks – what do you have?” At the top of Young’s opinion in AAUP v. Rubio, which ruled that Trump’s effort to deport foreign-born student protesters was unconstitutional.

October 8, 2025

The deployment of AI agents by enterprises is inevitable, and so are new power structures and manipulation risks. The underlying models will be the same for all users, but this possibility of divergence needs monitoring. Unlike traditional human oversight, no guardrails are in place for AI agents accessing sensitive materials with minimal checks.

So, are we advancing our systems through AI agents or surrendering agency before proper protocols are in place?

AI agents may learn and adapt quickly, but they lack the maturity gained from years of experience. Giving them autonomy with minimal supervision is like handing a company’s keys to an intoxicated graduate—enthusiastic, intelligent, malleable but erratic and in need of close monitoring.

What large enterprises fail to recognize is that this is exactly what they are doing. AI agents are “seamlessly” integrated into operations with little more than a demo and disclaimer, no continuous testing, and no clear exit strategy when something goes wrong.

A structured, multi-layered verification framework—testing agent behavior in simulations of real-world and high-stakes scenarios—is crucial as adoption accelerates. Different levels of verification are required based on the sophistication of the agent. Simple knowledge extraction agents or those trained to use tools like Excel or email may not require the same rigorous testing as more sophisticated ones that replicate a wide range of tasks humans perform.

When agents start making decisions at scale, the margin for error shrinks rapidly. If AI agents controlling critical operations are not tested for integrity, accuracy, and safety, we risk enabling them to wreak havoc on society. The consequences will be very real—potentially staggering costs for damage control.