Share with your CISO
Enterprise AI agents have become the fastest-growing attack surface in corporate environments, according to the Sophos AI Security 2026 Report, published July 22. The core finding is structural: AI agents routinely receive privileged access to core systems, and the OAuth tokens, service credentials, and API connections they depend on are now high-value targets. BeyondTrust data cited in the report puts active AI agent growth at 466.7% in a single year. Security governance has not come close to keeping pace.
What this means for your business
The exposure here follows a pattern that repeats every time a new class of identity proliferates faster than identity governance can absorb it. The question for any security leader right now is whether your organization’s AI agents are provisioned more like shadow IT or more like managed endpoints. If the answer is the former, even partially, you’re running privileged access on trust models designed for a world where identities were human, countable, and slow-moving.
Sophos is a security vendor with an obvious commercial interest in making AI threats feel urgent, but that incentive toward alarm doesn’t make the underlying dynamic wrong. The 466.7% agent growth figure, sourced from BeyondTrust rather than Sophos’s own telemetry, is the number worth sitting with. A nearly fivefold increase in a year means most organizations provisioned the majority of their current AI agent identities inside a window when their governance processes weren’t designed to handle them. The credentials issued during that window are the exposure.
The secondary threat the report surfaces is subtler and harder to instrument: an attacker who gains access to an AI agent doesn’t have to exfiltrate data immediately. They can steer the agent’s outputs gradually, poisoning the decisions it informs before any alert fires. That’s a different threat model than credential theft, and it won’t show up on a dashboard watching for unusual data movement. The organizations that should worry most are those where AI agents have already been woven into procurement, financial modeling, or customer-facing workflows, because the blast radius of a quietly manipulated agent is proportional to how much the business has already learned to trust it.
Concept deep-dive: AI identity
An AI identity is the set of credentials, tokens, and access permissions assigned to an AI agent so it can act on behalf of a user or system. Unlike a human employee, an agent can hold dozens of these connections simultaneously, operate continuously, and acquire new access programmatically without a human approving each step. The business risk is that existing identity governance tools were built to manage people, not software actors that scale horizontally and never clock out.
Based on reporting from AI Agents Now the Enterprises Fastest Growing Exposed Attack Surface, originally published 2026-07-23 07:30:00.
