Shadow AI incidents now account for 43% of all security breaches, up from 20% in 2025. The IBM Cost of a Data Breach Report 2026, which analyzed 602 organizations across 17 countries, ties that doubling to a record-high global average breach cost of $4.99 million—a 12% year-over-year increase. The U.S. average hit $11.5 million, the highest of any country and more than double the global figure.
The most expensive breach types are agent-native. Model inversion attacks cost an average of $6.07 million per incident. Prompt injection averaged $5.89 million. Both exploit the permissions and interactions of autonomous AI systems—precisely the layer that non-human identity governance is supposed to secure. The data correlates the lack of NHI governance with higher financial exposure: 92% of organizations that suffered an AI-related breach lacked proper AI access controls.
Cyera’s billion-dollar acquisition of Oasis Security reflects the market’s valuation of that gap. Oasis builds agentic access management for non-human identities—AI agents and automated software. The deal, the first $1 billion M&A in agent identity security, prices the governance necessity that IBM’s data quantifies from the breach side.
The scale is not niche. One in four organizations experienced an AI-driven breach in the past year. These attacks concentrate on critical infrastructure—62% of AI-driven incidents targeted that sector, with financial services and energy bearing the heaviest costs. AI-enabled breaches averaged $6.0 million, roughly $1 million above the global baseline.
“When adversaries can automate reconnaissance, generate persuasive phishing content, adapt malware and test exploits at machine speed, the cost and complexity of launching sophisticated attacks drops materially,” said Limor Kessem, Global Lead of X-Force Cyber Crisis Management at IBM. AI-driven attacks increased 56% year-over-year. Deepfake and impersonation tactics accounted for 45% of those incidents, AI-generated malware 19%, AI-generated phishing 17%.
The regulatory environment is not keeping pace. The EU AI Act’s high-risk obligations face a 16-month deferral to December 2027. Only 9 of 27 member states have designated competent authorities to oversee compliance. While the Nvidia Open Secure AI Alliance has recruited 37 members to build defensive infrastructure, and vulnerability discovery velocity continues to outstrip remediation capacity, the governance gap between AI adoption and AI security widens each quarter.
The defensive side is not barren. Organizations using AI and automation in security operations saved an average of $1.93 million per breach. Eighty-five percent of organizations aware of frontier models are increasing security spending. But the trajectory favors the attacker: IBM cites expert estimates that AI will provide a 31.7% advantage to attackers over defenders within two years. Only one-third of organizations have strict approval processes for deploying AI tools. Sixty-eight percent of breached organizations had no AI governance framework at all.
Shadow AI’s doubling from 20% to 43% in a single year is not a compliance footnote. It is the structural signal that AI adoption has outpaced the security frameworks meant to govern it—and the breach cost data now reflects that gap in dollars.