Key Takeaways

  • Cybersecurity Insiders reports that AI identities already operate inside key SaaS platforms with limited governance.
  • Researchers find steep visibility gaps, with 92% of leaders lacking full insight into AI accounts.
  • The rise of non-human identities intersects with broader insider risk trends documented by multiple industry studies.

Cybersecurity Insiders, working with Saviynt, has released new findings that illustrate a fast-developing exposure point for enterprises. The study, published April 21, 2026, outlines how AI identities are interacting with business systems in ways that mimic human users yet follow very different operational patterns. It is an unusual moment in security, because many organizations did not plan for machine-driven access at this scale. The result is a growing mismatch between system permissions and the teams responsible for tracking them.

According to the research, 71% of CISOs and senior security leaders confirm that AI tools already have access to core platforms like Salesforce and SAP. Only 16% report that those access rights are governed effectively. It is one thing to grant a service account limited privileges, but another when autonomous software can invoke APIs, hang onto long-lived credentials, or perform tasks that blur the boundary between workflow automation and unsupervised system behavior.

The visibility gap is striking. The survey reports that 92% of organizations lack full insight into AI identities that operate inside their environments. Even more concerning, 95% of leaders are not confident they could detect or contain misuse if something went wrong. That has become a recurring theme across the broader insider risk landscape. Recent data from the 2026 Cost of Insider Risks Global Report, produced by the Ponemon Institute, shows insider-driven incidents now costing organizations an average of $19.5 million per year. Negligent insiders, whether human or machine-based, account for $10.3 million of that total. The Ponemon Institute has been tracking these patterns for years, and the upward trajectory aligns with the rapid adoption of automation in cloud systems.

Enterprises are actively interpreting how non-human identities fit into existing control frameworks. A human employee can be trained, onboarded, and monitored. An autonomous agent, however, is often embedded inside a SaaS workflow, created by a business unit, or connected to a third-party integration with little centralized oversight. The founder of Cybersecurity Insiders put the issue plainly, noting that AI already accesses business-critical systems with more autonomy and less oversight than most security teams would knowingly approve. That is not an abstract concern; 75% of surveyed organizations have already discovered unsanctioned AI tools running inside their environments.

Industry observers have seen similar patterns emerge as organizations expand their cloud ecosystems. MIT Sloan researchers, for example, have documented the introduction of more than 170 new data protection laws in 2023 and 2024. Their work highlights how increased regulatory expectations are pushing enterprises to clarify which systems touch personal data and who, or what, can access it. In this environment, untracked AI identities can introduce compliance ambiguity that catches companies off guard.

A separate thread of context comes from the 2024 Insider Threat Report by Cybersecurity Insiders, which found a 48% increase in insider attacks year over year. As summarized by security research site StationX, the report emphasizes that insider risk programs often focus on human behavior, even as machine-driven operations become intertwined with sensitive data paths. Although the 2024 edition centered on traditional insiders, it provides a useful backdrop for understanding how new identity types compound older risks.

Broader cybersecurity trend reporting from analysts at Swif.ai points out how hybrid human-machine workflows create decision blind spots inside companies. While their focus is still primarily on user behavior, their work helps illustrate why organizations often underestimate the complexity of emerging identity ecosystems.

Returning to the Cybersecurity Insiders and Saviynt study, findings show that only 5% of security leaders feel fully confident they could contain a compromised AI agent. This highlights how detection tooling often assumes a human-driven threat model. Solutions from providers such as DTEX Systems, Proofpoint, and Securonix are starting to blend behavioral analytics with access governance, yet many of these tools were originally designed around human workflows. The shift toward machine identity monitoring is still underway.

What does this mean for day-to-day operations? Organizations are beginning to realize that autonomous agents behave more like distributed microservices than traditional accounts. They might operate continuously, trigger downstream actions, or interact with data in small bursts that appear benign in isolation. That said, the rapid adoption of generative AI inside business applications adds another layer. Business units sometimes deploy AI features without fully understanding their credential storage practices or the scope of their API permissions. It only takes one neglected identity to open an unexpected exposure path.

The study's conclusion is straightforward. As AI becomes more rooted in SaaS and cloud workflows, CISOs will need to prioritize continuous discovery, classification, and monitoring of machine identities. This shift aligns with guidance from standards organizations such as NIST, whose privacy framework encourages organizations to treat all identity types consistently from an access-governance perspective. Some teams are already experimenting with automated credential rotation, risk scoring for machine accounts, and enhanced approval workflows for non-human identity creation.

It is worth considering how rapidly this landscape could evolve. A year from now, the number of AI-driven workflows inside enterprise systems will likely be higher. The key question is whether governance strategies will catch up. Without better visibility, organizations may find that the next wave of insider incidents comes not from malicious employees, but from AI agents acting beyond their intended scope.