Big Tech’s AI race turns to who controls the agents

Microsoft, ServiceNow and Alphabet are making governance a selling point as businesses deploy more autonomous AI. A Coinbase supplier incident highlights the risks that can persist beyond a company’s own systems.

As big tech’s latest earnings season has shown, the enterprise AI conversation is shifting from capability to control. Microsoft, ServiceNow and Alphabet all used their recent results to highlight the security, privacy and compliance questions that come with deploying increasingly autonomous AI agents, while a separate disclosure from Coinbase offered a reminder that third-party suppliers can create customer-facing risk even when a company’s own systems are not breached.

Microsoft framed the issue as one of separating a model’s intelligence from its power to act. In the company’s first-quarter fiscal 2027 earnings call, chief executive Satya Nadella said Microsoft’s approach is to keep identity, permissions, monitoring and containment outside the model itself. The company also said customers used Microsoft Purview to audit more than 7.6 billion agent activities, a figure that underscores how quickly oversight demands are rising as AI tools are embedded into customer service, sales and back-office workflows. The practical challenge for enterprises is no longer simply whether an agent can reach a system, but whether it is authorised to do the specific task in front of it.

ServiceNow struck a similar note, but with a stronger emphasis on governance as an operating layer for enterprise AI. In its third-quarter results, the company positioned AI Passport and its Super Intelligence Control Tower as tools to help organisations discover agents, map their permissions, monitor behaviour and enforce restrictions in real time. Chief executive Bill McDermott described the moment as a “rules and rails” phase for enterprise intelligence, reflecting a broader industry view that security teams need a complete inventory of what agents exist, what data they can see and how they can be stopped if they misbehave. ServiceNow also unveiled a Control Tower Alliance involving technology partners including Anthropic, AWS, Cisco, NVIDIA, Okta and OpenAI, signalling that governance will increasingly depend on interoperability across vendors rather than controls inside a single platform.

Alphabet’s earnings discussion pointed in the same direction, with security requirements becoming more pressing as AI systems take on more complex tasks. The company’s third-quarter results also showed the commercial stakes: Google Cloud revenue rose sharply to $30.7 billion, helped by demand for enterprise AI infrastructure, AI solutions and core cloud services. That growth helps explain why security has become such a central selling point as well as a risk issue, because the more businesses depend on cloud-based AI, the more they need confidence that access controls, monitoring and response processes can keep pace.

The Coinbase disclosure provided a different but closely related lesson. According to reporting on the incident, contact information belonging to a small subset of customers was exposed through analytics provider Amplitude, while Coinbase said its own systems were not accessed without authorisation and customer funds remained safe. Even so, the episode illustrates how external vendors can introduce privacy and phishing risks that sit outside a company’s direct perimeter. Taken together, the week’s results suggest that enterprise buyers are now being asked to manage not just AI performance, but the full chain of trust around it: who can touch customer data, what automated systems can do with it and how quickly an organisation can contain the damage when something goes wrong.

Source Reference Map

Inspired by headline at: ^[1]

Sources by paragraph:

Source: Fuse Wire Services