More companies let AI agents handle money, email, and support

In short: Businesses are giving AI agents more control over finances, email, and customer support, and experts are warning that the risks can rise as fast as the benefits.

What's going on

AI agents are software tools that can take actions for you, not just answer questions (like a junior assistant who can click buttons and move money). Companies are using them to reply to customers, sort and send emails, and help make financial decisions.

The concern is that these agents can make the wrong call at scale. In financial services, an agent might “hallucinate,” which means it confidently makes something up, and then gives bad advice or approves a poor transaction. Some can also be tricked or used by criminals for fraud, scams, or cyberattacks, especially if the agent can access accounts and sensitive data.

There are also wider, shared risks. Researchers warn about “herding,” where many agents built on similar AI models or data react the same way during stress, like a crowd running to the same exit at once. In markets, that could worsen events like flash crashes or even speed up bank runs. Another worry is a single point of failure, meaning many businesses depend on the same AI provider, so one outage or breach could ripple across lots of companies.

What to watch

Firms and regulators are likely to push for stronger guardrails, better audit logs (records of what the agent did), and human review for high-risk actions like moving money or changing customer accounts. Deloitte and security guidance both suggest traditional controls may need updates when software acts with real autonomy.

Source: NYTimes