AI Agent Security Gap: 54% of Enterprises Hit by Incidents

What happened

A new wave of research from VentureBeat Pulse, covering 107 enterprises, has found that AI agents are being deployed with meaningful access to live systems and sensitive data well before the security controls needed to govern them are in place. The study identifies what researchers are calling an "agent security gap" — autonomous agents proliferating faster than the identity, isolation, and enforcement frameworks designed to contain them.

The findings reveal that more than half of the enterprises surveyed have already experienced a confirmed AI agent security incident or a near-miss. Despite this, most organisations continue to allow agents to share credentials rather than operate under individual, scoped identities — only around one in three enterprises assigns every agent its own distinct identity. More striking still, just three in ten isolate their highest-risk agents from one another or from broader systems.

The security tooling enterprises rely on is largely borrowed from model providers and hyperscalers rather than purpose-built for agentic architectures. Dedicated agent security spending remains a thin slice of overall security budgets, and enterprises are evenly divided on whether their current defences are keeping pace with AI-enabled attackers.

Why it matters

For customer experience and service-design leaders, this research lands as a direct operational warning. AI agents are no longer experimental — they are being embedded into customer-facing workflows, handling enquiries, processing transactions, and accessing personal data at scale. When those agents share credentials or operate without isolation, a single compromise can cascade across every touchpoint they serve, exposing customers to data loss, erroneous decisions, and broken trust that is extraordinarily difficult to rebuild.

From a behavioural economics perspective, the pattern here is a textbook case of optimism bias compounded by present-focus: organisations are capturing the near-term productivity gains of agentic AI while systematically discounting the probability and severity of downstream failure. Customers, meanwhile, have no visibility into the security posture of the agents serving them — meaning the reputational damage, when incidents occur, arrives as a shock rather than a managed risk.

By the numbers

• 54% of enterprises surveyed have experienced a confirmed AI agent security incident or a near-miss.
• 107 enterprises were included in the VentureBeat Pulse Research wave.
• ~1 in 3 enterprises assigns every agent its own scoped, individual identity.
• 3 in 10 enterprises isolate their highest-risk agents from other systems or agents.
• 50/50 split among respondents on whether current defences are keeping pace with AI-enabled attackers.

The Renascence take

Most of the conversation around this research will focus on cybersecurity and IT governance — which is understandable but misses the deeper customer experience exposure. The real risk is not just a data breach; it is the silent erosion of service integrity that happens when an agent operating with excessive privilege makes a wrong call on a customer's behalf and no one catches it in time.

The agent security gap is, at its core, a service-design failure disguised as a technology problem. When enterprises deploy agents without scoped identities or isolation, they are effectively removing the guardrails that make consistent, trustworthy customer interactions possible. Behavioural economics tells us that customers who experience a single unexplained failure by an automated system dramatically downgrade their trust in the entire brand — not just the channel. Customer-obsessed operators should treat agent identity and isolation not as an IT checkbox but as a foundational element of their service promise: if you cannot tell a customer exactly what your agent can and cannot access on their behalf, you are not ready to deploy it.