AI's New Power Brokers Leave Accountability Behind

At the 2026 IMF and World Bank Annual Meetings in Bangkok, officials weighed AI’s economic promise against the legal and administrative rules needed to control it. Cybersecurity experts warned that the UAE faces more than 800,000 attempted attacks a day. The machinery is advancing faster than the systems meant to oversee it. Citizens may be left asking who answers when an automated decision goes wrong.

Who Gets the Gains, Who Carries the Risk

Igor Ujhazi, Global Regulatory Technology Specialist and author of the Times of Israel blog post, attended the meetings as head of a firm working across public digital systems, regulation and market access. He said AI’s rapid development had outpaced the legal and administrative foundations needed to govern it, shaping decisions by finance ministers, central bankers and officials responsible for daily administration.

Figures presented in Bangkok forecast global expansion of 3.1 percent in 2026, with a modest rise expected the following year, while price pressures were projected to ease gradually. Ujhazi wrote that debt-service costs in lower-income countries claimed roughly 14 percent of government revenue. Meanwhile, large volumes of capital flowed into computing hardware, data centers and electricity generation before anyone could measure productivity gains with confidence.

Ujhazi said technology suppliers concentrated the benefits while risks spread more widely, particularly to energy importers facing higher costs and disrupted supply routes. Public institutions may not invest directly in these systems, but governments can still feel the effects through tax receipts, pension holdings and sovereign borrowing costs. Changes in labor demand could also reshape the tax base.

The Fund’s financial stability report warned that optimistic assumptions about returns could trigger abrupt market corrections. Concentrated exposure, borrowing and closer ties between technology companies and the financial system could carry a correction beyond its starting point. Public administrations see potential for faster services, improved fraud detection and more careful use of limited resources, but Ujhazi described deferred projects, uncoordinated pilot schemes and tools used without proper oversight.

The Rules Question

Kristalina Georgieva compared AI’s pace to a racing car that needs protective barriers, Ujhazi wrote. He argued that safeguards require laws specifying who may introduce a system, which decisions need human review, how officials assess performance and how people harmed by errors can seek correction. Privacy, cybersecurity, data retention, vendor access and documentation for independent examination need rules too, he said. International principles, he added, can’t replace operational rules suited to each country’s institutions and capacities.

The accountability gap is ordinary and concrete. Civil servants may be unsure what they can approve; citizens may not know who is responsible when an automated decision goes wrong. Ujhazi said agencies that rely on inconsistent registries, incomplete transaction records or weak identity checks may struggle to assess automated recommendations. He pointed to interoperable registries, verified credentials, examinable records and systems able to withstand disruption as necessary groundwork.

In fiscal administration, Ujhazi said an invoice tied to an identifiable taxpayer and preserved as reliable evidence offers a stronger basis for analysis than an isolated spreadsheet or unverified submission. Before automated systems take on consequential public functions, he called for human review, secure records, continuity procedures and tested recovery arrangements. Network failures, cyber incidents and service interruptions, he said, are routine planning concerns.

Autonomous Systems, Expanding Exposure

In the UAE, the UAE Cyber Security Council reported more than 800,000 attempted cyberattacks a day. Daily breach attempts on UAE infrastructure ranged from 90,000 to 200,000 early in the year before climbing past 800,000. The Council logged 128 confirmed incidents involving ransomware, data leaks and website defacement. Positive Technologies put the UAE’s share at 35 per cent of attacks in the region.

“AI has accelerated the cyber attacks dramatically,” said Dr Mohamed Al Kuwaiti, head of the Council. Global losses from cyberattacks reached $10.5 trillion, or Dh38.6 trillion, in 2024; forecasts put the figure at $15.6 trillion, or Dh57.3 trillion, by 2029.

Security executives warned that autonomous AI agents are taking decisions, accessing data and operating across company systems with little human oversight. “Every agent becomes a new identity, entry point, and target,” said Salah Suleiman, Managing Director, South Gulf at TrendAI. He said criminals use AI to launch “faster, more convincing, and scalable attacks.” TrendAI Vision One uses four principles—visibility, intent, control and oversight—and a dedicated AI security layer to govern models and agents.

Alexandre Depret-Bixio, SVP, International at Anomali, asked, “What do we allow machines to decide?” He said, “If you can’t trust your data, you can’t trust the decisions AI agents make on your behalf.” Daniel Churches, Director Cybersecurity, Asia Pacific and Japan at DXC Technology, cited DXC research showing only 47 per cent of organisations have fully integrated governance for agentic AI, even as many deploy autonomous tools. “Organisations are effectively scaling risk as fast as they’re scaling capability,” he said.

Justin Henkel, CISO at SolarWinds, said every automated actor needs “a defined purpose, bounded authority, trustworthy provenance, observable behaviour and a tested path to revocation.” Depret-Bixio urged patience: “Autonomy earns trust one governed decision at a time, with human approval and audit where it matters.”