By 2028, unmanaged AI could contribute to a quarter of enterprise breaches. Pressure from data sovereignty is also increasing

On September 23, at the Security & Risk Management Summit in London, Gartner presented seven forces that analysts say are reshaping companies’ security environment. They include uncontrolled proliferation of AI agents, fragmentation of technology-sovereignty requirements, the impact of autonomous AI on employees, diverging labor-market developments, changes in customer interactions, system opacity and new attacks conducted through circular IT supply chains. The most concrete forecast concerns agents. Gartner expects abuse of unmanaged AI agents to contribute to 25 percent of enterprise security breaches by 2028. An agent is more attractive to an attacker than an ordinary chat tool because it has permission to do something: read a database, work with documents, use an enterprise interface or change the state of another system. The risk is also growing from inside companies. Individual business units can create their own agents faster than security teams can inventory and control them. An organization may then not know exactly how many autonomous systems it operates, what they can access or who is accountable for them. Gartner therefore calls for zero-trust governance and technical kill switches assigned to individual agents. The second forecast has a direct legislative and geopolitical dimension. By 2028, Gartner says 30 percent of multinational organizations will be forced to redesign their AI architecture because of differing national sovereignty requirements. This is expected to raise their operating costs by 20 percent. The reason is not only personal-data protection. Governments are increasingly addressing where data may be stored, which infrastructure may process it, who controls the models being used and whether sensitive systems may depend on foreign providers. A multinational company may therefore discover that one global model and one shared cloud architecture cannot be used in every country. This weakens the economic advantage of centralization. A company may need regional model variants, separated data layers, multiple providers and different control mechanisms. Each such variation increases the cost of integration, testing, updates and administration. Security leaders therefore need better system observability. Gartner emphasizes that the ability to see what AI is actually doing is a prerequisite for trust and risk management. With a traditional program, most steps can be predetermined. An autonomous system can choose its own path to a goal, so the company must monitor its real behavior during operation. For company leadership, it is important to connect security, legal and financial planning. A decision about a model or cloud provider is no longer merely a technical choice. It can determine whether a service will still be legally and operationally usable in all countries where the business operates two years from now. AI therefore creates a peculiar paradox. Technology standardization can reduce costs in the short term, but growing geopolitical fragmentation may later require expensive decentralization. In long-term supplier contracts, the ability to move data, models and workflows between different environments is therefore becoming increasingly important. KEY TERMS Data sovereignty: A requirement that the storage and processing of data be subject to the rules of a specific country or territory. Zero trust: A security principle under which no user or system is trusted automatically and every operation is verified. AI observability: The ability to monitor a system’s steps, decisions, consumption and behavior during real operation.

By 2028, unmanaged AI could contribute to a quarter of enterprise breaches. Pressure from data sovereignty is also increasing

The full article is available to users registered at Hard Skills