The security risks posed by AI agents dominated discussions this week. GitGuardian’s report exposed 23.7 million leaked secrets on public GitHub repositories in 2024, driven largely by unmanaged non-human identities (NHIs) like AI agents. With companies managing 45 machine identities per human employee, credentials like API keys often go unchecked, creating loopholes for hackers. Tools like GitHub Copilot worsened the problem, increasing leaks by 40% in enabled repositories. Experts warn that poor NHI governance could lead to long-term vulnerabilities as AI agents proliferate.

Governance solutions are emerging to address trust gaps. SAS launched its AI agent framework in Viya, emphasizing transparent decision-making and embedded ethics. The platform allows organizations to customize how much autonomy AI agents have, ensuring humans can review critical choices. This "ethical calibration" aims to prevent AI from acting unpredictably or beyond its intended scope. Analysts like Nick Patience of The Futurum Group praised the approach for balancing innovation with accountability.

The debate over supervision vs. autonomy intensified. Salesforce’s AI agents already handle customer service tasks independently but are programmed to seek human help when uncertain. This hybrid model is seen as a blueprint for safer deployments. However, proposed personal AI agents—which could negotiate purchases or manage communications—raise thorny questions. Who controls these agents’ priorities? Could their loyalty be split between users, advertisers, or developers?

Adoption trends suggest a pivotal year for agentic AI. An IBM-Morning Consult survey found 99% of developers actively building AI agents, with many targeting enterprise use cases like data analysis and workflow automation. Despite enthusiasm, real-world agents still struggle with complex tasks, highlighting a gap between hype and reality. Analysts stress that trust will depend on proving reliability through auditable systems and clear boundaries for AI decision-making.

Globally, industries like healthcare are prioritizing human-centered AI design to foster trust. While not directly tied to agents this week, the push for transparency in medical AI reflects broader demands for accountability—a lesson agent developers are urged to adopt.