Dive Brief:
- Interest in AI security training rose dramatically this year as businesses grapple with the complex task of securing systems and data against vulnerabilities, according to a Coursera report published Tuesday. The online learning platform analyzed data from more than 300 million enterprise and consumer course participants across Coursera and learning platform Udemy, which it acquired last year.
- Participants in AI security training rose 665% globally, while U.S. enrollment increased 358% in the same time frame. By comparison, global enrollments in all AI related courses grew 67% from last year.
- In a sign of broader enterprise deployment, training around agentic AI workflows also grew significantly across the world, up more than sixfold from last year, according to the report.
Dive Insight:
Broader agentic AI deployment in enterprise settings increases the potential risks of the technology, as agents gain access to critical data and become more entwined with operations. Externally, risks are also on the rise as more sophisticated models learn to break through their safeguards and attack organizations.
For CIOs, the task of ensuring secure AI deployment hinges on a human component, as lagging skills can slow down enterprise AI plans. Almost 4 in 10 businesses lack the skills needed to secure and govern AI tools in use across operations, according to a Barracuda report.
The risks of AI continue to emerge and include a string of breaches at companies and governments executed by frontier AI models.
“We’ve seen a number of AI models escape their sandboxes this year and attempt to, or successfully, hack other companies,” said Shaila Rana, a cybersecurity professor at Purdue Global and senior member of IEEE, in an email to CIO Dive. “This year has shown that we can’t rely on the agent itself to behave. We have to build the architecture around it.”
Rana said companies must apply zero trust principles around agentic AI deployment, ensuring every action is verified, every permission is scoped and the blast radius is contained if something goes wrong.
"Human in the loop was designed for a pace of decision-making that agents have outgrown,” Rana said. “The shift is toward a human above the loop, where people set the boundaries, monitor behavior across the system and step in when an agent does something it shouldn't, rather than approving each individual step.”
