AI Agents With Wallets Need Their Own ID Layer, Says BGA Report Backed by Deloitte and ADGM

In Brief:

• April 2026 saw the Blockchain Game Alliance release “Agentic AI in Gaming and the Emerging Digital Economy,” a paper whose contributors included Deloitte, Abu Dhabi Global Market and Teranode Group.
• Two governance ideas anchor Teranode’s section — Know Your Agent and an AI Control Tower — each built for agents that control wallets and shift funds without a human signing off on every transaction.
• That research became the foundation for the AI & Agentics Council, launched in late July with Deloitte and DMCC, while Teranode takes the Control Tower to Gamescom in Cologne on Aug. 26.

Before studios hand AI agents access to player assets or treasury money, the Blockchain Game Alliance says those agents need an identity layer of their own — one distinct from KYC and KYB.

The argument is laid out in “Agentic AI in Gaming and the Emerging Digital Economy,” a six-section report the alliance put out in April 2026 alongside contributors including Deloitte, Abu Dhabi Global Market and Teranode Group. What the document asks for is machine identity capable of handling authentication, authorisation, traceability and policy enforcement where the actor is automated.

Promoting the paper, the alliance wrote that “the next 18 to 36 months will fundamentally reshape the digital economy,” a reference to agents leaving sandboxes behind for live, persistent virtual economies where real capital is at stake.

Know Your Agent

Two concepts came from Teranode Group. Know Your Agent, or KYA, is the first, proposed “because future financial systems may need to verify not only the customer, but also the agent acting on their behalf,” according to Teranode Group’s Thomas Giacomo. The second is an AI Control Tower, resting on the premise that “governance must become continuous, policy-driven and real-time.”

Giacomo put the change in plain terms: “We are moving from AI systems that generate content to AI systems that execute actions,” he said.

On the technical side, Teranode combines hardware-based secret management with runtime attestation so every agent traces back to an accountable person or organisation. The company noted that tougher machine identity won’t turn an agent into a legal counterparty — what it does is make someone answerable for the agent’s behaviour.

Teranode contributors credited on the report are Aurélien Munoz, Robert Alizon, Liuxuan Pan, Marcin Pakielo, Pieter van Eck and Rafa Jiménez.

How the Control Tower works

Munoz, lead product manager at TeraNode Group, calls the Control Tower “an unpassable Digital Toll Booth,” positioned between what an agent intends to do and the API call that would carry it out.

When a stablecoin is about to leave a wallet, the tower verifies the agent’s identity, its transaction limits, and whether the requested action lines up with the human user’s original instruction. Anything above a safety threshold is either blocked outright or handed to a person for manual approval.

Regulation enters via the EU AI Act, under which financial systems — and the AI agents running them — count as high-risk. Rather than letting code operate unsupervised, Munoz said studios should pursue governed autonomy, combining KYA protocols with pre-execution control.

Standards and cost

The report also charts the standards stack these agents would sit on: ERC-8004, live since January 2026, for identity and reputation; x402 for HTTP-native payments; ERC-8183, still in draft, for agentic commerce; plus the A2A and MCP communication protocols.

Using GPU count to size infrastructure gets pushed back on too. According to the report, costs stem from idle accelerators, repeated retrieval, duplicated embeddings, overprovisioned storage, tool-call overhead and agent loops that aren’t properly controlled — and the only metric worth tracking is end-to-end cost per useful task.

As the report frames it, three features set blockchain games apart from Web2 free-to-play: agent-level reputation and identity, direct wallet management and a token-based economy.

From report to council

April’s research paved the way for the AI & Agentics Council, which the alliance introduced in late July together with Deloitte and Dubai free zone DMCC. Xsolla, SKALE, The Sandbox and WAM are the founding partners.

Structured as a department within the BGA, its scope takes in governed autonomy, human accountability, model-independent infrastructure and economies in which agents take part alongside players. What is holding institutional money back, the council says, is the still-unresolved liability question: who answers when an agent carries out a transaction by itself.

Founded in 2018, the BGA has roughly 350 members and has been led since 2020 by president Sébastien Borget, co-founder and COO of The Sandbox. Borget observed in the report that gaming had AI agents well before large language models arrived, with earlier versions built on reinforcement learning in which agents learned through rewards or penalties.

The AI Control Tower demo lands Aug. 26 at Cologne’s Agent AI Summit, subtitled “Architecting Autonomous Worlds.” ONEchain is in the same city that day with an AI-native gaming economy of autonomous agents that “compete, create, trade and earn on-chain.”