UNITED KINGDOM / AGILITYPR.NEWS / June 12, 2026 / There is a line in IBM training material from 1979 that has resurfaced with uncomfortable relevance in the age of autonomous systems: “A computer can never be held accountable, therefore a computer must never make a management decision.”

For decades, it read like a philosophical warning. In 2026, it increasingly reads like an unresolved design flaw.

AI consultancy ResearchCollab Technologies has warned that organisations deploying agent-based systems are now confronting an accountability gap that traditional governance models were never built to handle.

The issue is not that AI is making mistakes in isolation. It is that decision-making is being distributed across systems that no longer behave like tools in the conventional sense. They operate, delegate, trigger other agents, and execute actions across environments without a single clearly accountable human decision point.

Recent research from Ping Identity and KuppingerCole Analysts notes that AI agents are already being deployed into production faster than enterprises can govern them. That imbalance is exposing identity and access systems designed around human users rather than continuously operating digital actors.

Findings from IBM's 2025 Cost of a Data Breach report show that 13% of organisations have already experienced AI-related security breaches, and a staggering 97% “lack adequate access controls” for AI systems.

Two concepts in particular are beginning to define the problem: delegation opacity and sub-agent spawning. In practice, this means organisations may not be able to reconstruct how a decision was produced once multiple agents interact, inherit permissions, and trigger downstream actions across systems.

It is not a fringe scenario. It is already occurring inside enterprise environments where identity systems struggle to maintain traceability once actions move beyond single-user authentication flows.

Deloitte’s State of AI in the Enterprise report adds a blunt measure of readiness: only 21% of organisations have reached a mature state of AI governance. The majority are therefore scaling systems whose decision pathways they cannot fully define, let alone govern consistently.

Imran Chughtai, Founder and Chief Executive Officer of ResearchCollabTech, said, “Most governance models assume decisions can be traced back to a human actor. That assumption is breaking. When agents begin delegating to other agents and executing in real time across systems, accountability becomes fragmented. If no one can clearly own the decision chain, then governance becomes procedural rather than real.”

ResearchCollabTech works with organisations to address this gap by helping define accountability structures that extend across human teams and autonomous systems. That includes assigning explicit ownership for agent behaviour, defining escalation paths that remain enforceable at runtime, and embedding oversight into the operational layer of AI systems rather than treating it as documentation after deployment.

The consultancy argues that most organisations are still applying governance frameworks designed for deterministic software environments, where outputs are predictable and system behaviour is stable. Agentic AI does not behave this way. It introduces emergent workflows, conditional delegation, and cross-system actions that evolve in ways traditional audit models were never designed to track.

The central question is no longer whether AI systems can act autonomously. It is whether organisations can still answer a basic governance question when they do: who is responsible for what just happened?

For more information, visit researchcollabtech.com.