AI agents are spreading fast. Their rules are still catching up.

AI agents are rapidly spreading across enterprises, but the rules governing their control have not kept pace. A recent survey found that 86% of respondents are already using AI agents embedded in applications, yet only 12% feel their organizations fully understand the risks addressed by sovereign AI. These agents can now interact directly with company systems and data, creating significant challenges in terms of observability and guardrails around access to sensitive information.

Cohere, a Canadian AI company, conducted the survey, emphasizing that control should be built into the agent architecture. Cohere's chief AI officer, Joelle Pineau, states that sovereignty is an architectural issue. If an AI system is tied to a single provider or governed by terms that can change, it lacks control and becomes dependent.

The survey, which included 508 IT and business decision-makers from large organizations with $1 billion in annual revenue across four countries, revealed that only 13% of respondents felt sovereign AI concepts were widely understood. Despite this, AI adoption is high, with 99% of respondents using generative AI and 86% using agent-enabled applications.

The survey predicts that agent adoption will further increase, with 67% of respondents expecting to use prebuilt or third-party agents within the next year. Cohere's solution involves running the complete platform, including models and agents, on the customer's own infrastructure. This setup allows for full control over model access, updates, and patches, but also shifts responsibility for management back to the enterprise.

A major barrier to sovereign AI readiness cited by respondents is infrastructure, followed by skills and talent, cost, and budget concerns.