Opens in a new tab
vmblog logo 2024 wht (updated)

As AI Gains Autonomy, Enterprises Must Tighten Control in 2027

Share: 

Joe Logan | Published: November 17, 2026
vmblog predictions 2027

By Joe Logan, CIO at iManage

In 2027, the challenge of AI will move beyond what the technology can do to how it’s governed, managed, and paid for. As AI agents proliferate, their growing autonomy will raise questions about identity, data access and accountability.  At the same time, the quality and consistency of enterprise information will determine whether AI supports sound business decisions or amplifies existing problems. CIOs will need to find the happy medium to balance risks with AI costs, reassessing their technology portfolios and establishing clear controls to ensure AI delivers value, not complexity.

Here are my opinions on the key areas enterprises will need to focus on in 2027:

AI agent identity management crisis looms

As AI agents proliferate, identity management is one of the key challenges CIOs face in 2027. Without clear identities, tightly defined permissions and visibility into agent activity, enterprises risk exposing confidential HR, legal or financial data, creating compliance headaches and undermining trust in AI-driven operation.

The issue is no longer simply what agents can access, but what their authorized scope is, and how data access permissions are managed, especially as agents begin spawning other agents, inheriting each other’s identities.  

Agents operating beyond their intended scope, accessing and exposing information due to misconfigured and mis-secured data repositories, present a huge risk to enterprises. Continuous auditing of underlying data sources and fixing unauthorized agent access will form a significant focus for CIOs.   

AI agent deployment success will mean more than providing an agent to complete a task. It’ll mean routinely verifying that agents stay within their defined remit, only doing what they are designed to do.

All this requires clear process governance, discoverable agent identities, time-limited data access, auditable activity and rigorous testing. As agentic AI scales, the defining measure of success may well be how effectively enterprises control what their agents cannot do.

Data governance failure could be the next enterprise AI scandal

As AI models improve, hallucinations will become less of a concern, with the bigger risk being agents retrieving information that is real but wrong, outdated or poorly governed.  The problem always isn’t what AI makes up, but the quality of data it’s allowed to use and how that information influences business decisions.

Enterprises need strong data governance and clear business context. They need a shared dictionary of terms, with each business definition linked to a specific technical definition so that both people and AI understand the data in the same way.

Likewise, knowledge management must connect definitions across systems, departments and teams. Instead of relying on siloed or conflicting meanings, enterprises will need to provide AI with one consistent, complete context so that AI can interpret and use the information accurately.

AI budgets move from individual team spend to organizational expense

AI agents become an enterprise cost centre

As AI agents become institutionalized, their costs will shift from individual team budgets to wider enterprise spending. To optimize costs, organizations will need to monitor whether the frequency of agents running is appropriate, their scope is clearly defined and how efficiently they operate.

Testing will help validate if their outputs are identical every time the same task is completed, but also whether the tool or agent can be adjusted to improve efficiency and provide a clearer understanding of cost-to-value.

Utilizing controlling servers or brokers to see the whole picture will become imperative. This will help report on the agents that are running, what they cost and the value they provide. 

Tool consolidation and cost trade-offs

Reassessing the broader technology portfolio as AI becomes capable of replacing some existing tools will become an enterprise priority. Specialized products that provide meaningful, differentiated benefits will continue to warrant investment, while other paid tools may become obsolete and generate savings. However, enterprises must be watchful, as the savings from the outdated tools could be offset by inefficient AI deployments and a proliferation of AI tools that deliver limited value – ultimately increasing the overall cost of technology tools. 

Vendors who incorporate AI as a value-add into standard offerings and raise prices will need to rethink their strategy. Creating new cost pressures even as customers may not need every newly added capability may prove self-defeating. 

In summary, the next phase of enterprise AI will be defined by how effectively agents are controlled. Clear identities, appropriate permissions, and reliable and consistent data will be essential. Before acquiring new AI capabilities, their expense will need to be justified, just as existing tools will need to justify their place in the technology stack. Successful AI adoption will depend greatly on disciplined governance and not merely enthusiasm for the technology itself.