By Thordis Thornsteins
In 2026, agentic AI will pull businesses back into an experimental phase, where the risks rise as fast as the opportunities.
The early days of GenAI resembled a tech ‘Wild West’. Organizations experimented with AI without fully understanding its limitations, resulting in frequent errors, such as engineers giving valuable source code to ChatGPT, essentially leaking IP to the entire world. Over time, however, organizations have brought much of this chaos under control through stronger governance, clearer policies, and more mature operational practices.
Agentic AI will open the gates again, shifting the risk landscape and raising the stakes even higher.
Because these systems act autonomously, there is even less oversight and greater potential for chaos to spiral out of control. Even minor issues, such as authentication faults or misconfigurations within the AI system or its dependent processes, could cascade across companies, exposing sensitive data or triggering unintended actions.
As organizations increasingly delegate decision-making to AI agents, these mistakes will be amplified, making proactive governance essential. Gaining complete visibility over the systems that AI interacts with, enforcing strict access controls, upskilling teams, and embedding robust governance frameworks will be essential to maintaining innovation while controlling risk, and avoiding a return to the bad old days of the AI Wild West.
The author
Thordis Thornsteins is Lead AI Data Scientist at Panaseer
