Can you trust your AI Agent?

Personal AI agents—software tools that can act independently on your behalf—are no longer futuristic. They’re being actively rolled out by companies like Salesforce and Spotify, performing tasks from customer support to personalised music curation. Soon, they’ll manage your calendar, negotiate purchases, and even send emails for you.

But here’s the catch: as these AI agents grow in autonomy, a critical question emerges—can they be trusted? Like human employees, AI agents will make decisions using your data, money, and preferences. That raises serious risks around security, manipulation, and misinformation.

Four key concerns stand out:

1. Criminal Interference: Agents could be hacked or manipulated to act against you.

2. Retail Bias: Some may push products or services from their developers or sponsors.

3. Advertiser Loyalty: Just like social platforms, AI agents could favour content that benefits advertisers over what’s best for you.

4. Misinformation: AI agents might make decisions based on false or misleading information.

To build trust, the authors propose treating AI agents like legal fiduciaries—those who are legally obligated to act in your best interests. This means clear rules, strong accountability, and tools for users to monitor or limit AI behaviour.

They also suggest market-driven safeguards like built-in insurance, independent audits, and new oversight services (like “AI credit bureaus”). Finally, tech companies must design AI agents that prioritise local processing and strong encryption to protect personal data.

The bottom line? Agentic AI has incredible potential, but without strong legal, technical, and ethical guardrails, it could backfire. The time to act is now—before these agents become deeply embedded in our digital lives.

Reference:
Levin, B., & Downes, L. (2025, May 26). Can AI Agents Be Trusted? Harvard Business Review.