AI Is Involved in 84% of Ransomware Attacks, but the Bigger Threat May Already Be Inside Companies

Cybercriminals are finding a new use for corporate AI tools: turning systems designed to help employees into another way to reach sensitive data.

The Semperis 2026 Ransomware Risk Report found that 84% of the ransomware victims who were surveyed experienced attacks that involved AI. Even more alarming is the fact that 44% of them said the attackers had used their own AI agents or systems. The results are based on 1,200 IT and security professionals from nine countries, all of whom work for organizations that were affected by ransomware the previous year. The findings summarize the experiences as reported by the victims, not a measure of all ransomware attacks worldwide.

AI has made the cybercrime techniques which are already established more efficient since it allows criminals to produce phishing messages, speed up the development of malicious code, and carry out vulnerability analyses with less technical knowledge.

Chris Inglis, former U.S. National Cyber Director, told CBS News that more autonomous attacks are already appearing. However, using AI to create ransomware differs from using an AI agent already connected to corporate systems.

Companies are giving agents access to internal documents, messaging platforms, databases, and operational tools. These links enable agents to carry out useful tasks, sometimes spanning multiple applications without human intervention. They also provide attackers with opportunities. Should criminals obtain the agent's credentials, alter its instructions, or take control of an authorized session, they could gain access to the applications and the data accessible through that agent. The weakness may not lie in the model itself; it may lie in the permissions associated with it.

Semperis found that 87% of the ransomware attacks surveyed targeted identity infrastructure, including Microsoft Active Directory, Entra ID, and Okta; these systems determine who can access corporate resources and what actions they can take after authenticating.

Because identity infrastructure can be compromised, it can give attackers broader privileges, which is more valuable than compromising a single application. But only about half of the organizations surveyed kept dedicated identity backups, and fewer than half had automated identity recovery.

For any company using AI agents, these gaps deserve special attention, since each agent with access to the company's systems creates another identity, a set of credentials, or a delegation of authorization that must be protected.

The findings suggest three immediate priorities for enterprise AI deployments:

• Limit agent permissions: Give agents access only to the systems and operations their tasks require.
• Enforce authorization independently: Sensitive actions should be checked against current policies rather than approved solely because an agent has valid credentials.
• Protect identity recovery: Maintain tested backups and recovery procedures for identity infrastructure, not just business data.

The report does not prove that AI caused the breaches surveyed or that most ransomware operations were fully autonomous; it does, however, show how extensively attackers are incorporating AI into their existing methods.

AI may accelerate ransomware attacks, but the more consequential failure would be allowing an attacker to turn permissions granted for legitimate automation into authority the organization never intended to give.