Enterprises Face Growing Data Security Pressure as AI Agents Enter Core Work
TechRadar reports that autonomous AI agents are moving into core business workflows, widening attack surfaces and forcing companies to rethink data security and trust.
According to TechRadar, the rapid spread of AI tools in business is changing what companies need from data security and the trust customers place in them. The outlet argues that autonomous, decision-making AI agents are moving beyond pilot projects and into core business workflows, expanding the enterprise attack surface about as quickly as they are being adopted.
TechRadar cites a Gartner prediction that 40 percent of enterprise applications will include task-specific AI agents by the end of 2026, up from less than 5 percent in 2025. The article reports that Gartner advises leaders of software organizations to define an agentic AI strategy soon, warning that companies that delay could fall behind competitors.
The outlet also says the security model itself is shifting. Encryption and perimeter defense, long the foundation of corporate protection, are described as no longer enough on their own as AI-driven activity moves through business systems. TechRadar does not spell out a full set of replacement controls, and specific tools or standards are not detailed in its coverage.
Taken together, the report frames AI agent adoption as a race in which security and governance must keep pace with deployment.
Background
Enterprise security has traditionally rested on two pillars. Encryption protects data while it moves across networks and while it sits in storage. Perimeter defenses, such as firewalls, separate internal systems from outside threats. Over time, identity and access management became a central third layer, deciding who or what can reach which resources.
AI agents differ from simple chatbots in an important way. A chatbot generates text for a person to read. An agent can call software tools, read and write records, trigger transactions, and chain several steps together, often with limited human review at each stage. That reach is what makes agents useful, and it is also what makes them a security concern. A single flawed instruction or compromised credential can propagate across connected systems more quickly than a human could intervene.
Security researchers have documented several categories of risk tied to AI systems. Prompt injection occurs when malicious instructions are hidden inside content an agent processes, such as an email or web page, and manipulate its behavior. Another common concern is over-permissioning, where automated accounts receive broader access than their tasks require. Established practices, including least-privilege access, detailed logging, and human approval for sensitive actions, are being adapted to cover these automated actors.
Analysts have long tracked the gap between experimental AI projects and production deployments. Moving into production forces harder questions about where data comes from, who is accountable for decisions, and how actions can be audited afterward. Gartner is a research firm whose forecasts are widely used in corporate planning. Such predictions are estimates rather than measurements, and their accuracy depends on how the underlying terms are defined.
Why it matters
The shift TechRadar describes affects more than engineering teams. When software begins acting on a company's behalf, the consequences of weak security reach finance, legal, customer service and compliance. Errors or breaches may involve customer records, internal strategy, or regulated information, and the damage can be harder to contain when automated processes move quickly.
The story also highlights a tension that many organizations already feel. Competitive pressure encourages rapid adoption of tools that promise efficiency. Caution, by contrast, requires time for policy design, testing and staff training. Companies that move fast without governance may gain short-term advantages while accumulating risk they do not fully see. Companies that move slowly may find that delay carries its own costs.
For security vendors and cloud providers, the trend suggests a growing market for tools that treat AI agents as identities requiring oversight, not just as software features. For customers, it raises the importance of asking how a product's automated components are monitored, restricted and reviewed. Trust, which TechRadar identifies as central, is likely to depend on whether organizations can explain what their agents did and why.
What to watch
The first question is whether the adoption forecast holds. Watch for independent surveys and company disclosures that show how many agents are truly deployed in production, and how many are better described as conventional automation with new branding.
Second, observe how identity and access platforms adapt. Changes to permission models, audit logging and approval workflows for non-human accounts would indicate that the industry is treating agents as a distinct category of risk.
Third, look for regulatory and standards activity. Guidance on accountability, data handling and disclosure for automated decision-making could shape how companies design their agent deployments over the coming year.
Finally, watch for reported incidents. Public accounts of agent-related security failures, and how organizations respond to them, will show whether current controls are sufficient or whether the field needs new practices.
Viewpoints
Commentary from our columns: three angles on this story.
The Skeptic open questions and weak points
Forecasts are useful, but they are not facts, and this one deserves a closer look before it steers budgets. A projection that 40 percent of enterprise applications will feature task-specific agents by the end of the year depends heavily on how "task-specific agent" is defined. Does a feature that drafts a reply count? Does a scheduled script with a language model attached count? The jump from under 5 percent in 2025 is steep enough that definitions matter a great deal. The urgency in the message is also worth examining. Warnings that competitors will pull ahead are persuasive, and persuasive language is not evidence. Leaders should ask which vendors and analysts benefit from a sense of rush, and whether their own internal data supports the same timeline. The security concerns raised here are real. The open question is how much of the risk is new and how much is familiar risk with a fresh label.
The Ledger the business angle
Every technology cycle forces a choice between spending now and spending later, and agentic AI sharpens that choice. The business case for adoption is easy to state: faster processes, lower costs, and fewer manual steps. The business case for security is less visible until something goes wrong, which makes it easy to underfund. Boards and finance teams should treat AI security as an operating expense with a clear return, not as an optional add-on. That means budgeting for access controls, monitoring, testing and staff training alongside the licenses themselves. Vendors have an incentive to bundle security into their platforms, and buyers should compare those offers closely. Companies that can show they adopted agents responsibly may win customer trust that becomes a competitive advantage. Those that cut corners may find the savings consumed by a single serious incident.
The Rulebook privacy, safety and regulation
When software begins acting on behalf of a company, the central legal question becomes accountability. If an automated agent accesses personal information, makes a decision, or shares data it should not have, who answers for that outcome? Organizations need clear answers before deployment, not after an incident. That requires documented purposes for each agent, limits on the data it can reach, records of what it did, and a path for humans to review and reverse its actions. Privacy principles such as data minimization and purpose limitation apply just as much to automated systems as to traditional databases. The regulatory landscape for AI is still developing, and companies should expect rules to tighten rather than loosen. The strongest protection for customers is not a single technical control but a culture that treats every automated actor as a responsibility with a name attached to it.