AEREDIUM Puts Limits on AI Agents That Move Money and Write Code.

Giving an AI agent access to a code repository or a wallet turns a software instruction into something with real consequences. The next question is who decides which actions it can take—and when a person needs to step in.

AEREDIUM is bringing that decision into a shared control system for software development and digital asset transactions. In an announcement dated October 22, the company outlined two products, The Chair and MCP Connector, which it says launched on October 13 through its new AI platform.

Both evaluate proposed actions against rules set by an organization or agent owner. Each request receives one of three outcomes: approved, denied or held for human review. Decisions are signed and recorded so the organization can later check what happened and which permissions applied.

The approach gives businesses a way to delegate routine work while keeping approval requirements attached to sensitive actions.

AEREDIUM positions the tools as an extension of its institutional infrastructure. In an October 20 post, founder Albert Dadon described the underlying blockchain’s division of responsibilities: software validates transactions, while identifiable people verify the resulting record.

𝕏Post from @aerediumView the original post on X↗

A Check Before Code Runs

The Chair focuses on AI coding agents working inside an organization’s development environment. According to AEREDIUM, it evaluates tool calls before execution, covering actions such as running commands, modifying files, merging code and pushing changes.

Existing policies can automatically approve or deny a request. Actions outside those rules go to a control panel, where an authorized person decides whether they should proceed.

For a development team, that could mean allowing an agent to make routine file changes while requiring a person to review a more consequential operation. The boundaries depend on the policies the organization establishes.

Each decision generates a signed receipt documenting the attempted action, the policy applied and who authorized it. That connects the agent’s work to an approval history, giving teams a record to examine when reviewing changes or investigating an incident.

The Chair supports Claude Code through a dedicated plugin and offers a governed shell for other coding agents, including those from OpenAI. Organizations can configure policies through a questionnaire rather than writing the rules directly in code. AEREDIUM is offering a 30-day free trial.

A Payment Agent Cannot Become a Trader

MCP Connector applies the same approval model to agents handling money. It combines a policy engine with a digital asset wallet, with each agent assigned its own wallet, policy and designated owner.

The owner defines the agent’s role, approves its mandate with a passkey and can halt or delete the agent. The product currently offers two roles with distinct permissions:

• Payer: Can send USDC, USDT, PYUSD and EURC to owner-approved addresses, but cannot trade.

• Trader: Can trade on designated exchanges within its own wallet, but cannot send funds to external recipients.

That separation matters because permission to perform one financial task does not automatically justify access to every other task a wallet supports. A business could authorize an agent to make approved payments without also giving it the ability to take trading positions.

AEREDIUM says MCP Connector supports Ethereum, Arbitrum and Base and retains transaction records for seven years. It is designed for Claude and is open to OpenAI and other AI frameworks.

The company says its wallet uses threshold signatures generated inside secure enclaves, with no complete private key created and no key held by AEREDIUM to the owner’s funds. Those are the company’s stated security characteristics; the controls described in the announcement govern access and execution rather than guaranteeing the outcome of an approved trade.

Why Banks Are Drawing the Line

The launch comes as banks draw tighter boundaries around outside agents. As Blockster reported on October 19, Citi and Bank of America have restricted AI agents’ access to customer accounts. Citi bars third-party agents from logging in or transacting on its digital platforms, while Bank of America has emphasized human authorization before money leaves an account.

AEREDIUM’s products address a central concern in that debate: how to give software useful authority while keeping its actions within enforceable limits. Its wallet-based approach lets owners define permitted activity in advance and send exceptions for human review. That is a separate setup from an outside agent accessing a bank account; it does not change either bank’s access rules.

Keeping Authority Attached to the Action

The two products extend AEREDIUM’s existing work in institutional digital asset infrastructure. The company says both draw on its policy engine, with its financial infrastructure also using threshold-signature technology to enforce rules before transactions are signed.

CEO and founder Albert Dadon framed the launch around the consequences of giving software more responsibility.

“The challenge isn't just making these agents more capable, but making sure they operate within the rules we set.”

— Albert Dadon, CEO and Founder, AEREDIUM

The launch fits a broader shift in agent infrastructure: linking an automated action to a person, a mandate and a record of authorization.

Blockster recently covered World’s move into human verification for banking, including tools that connect agents to verified people. AEREDIUM addresses the permissions that determine what an agent may do once it has access to a system.

A related question runs through t54’s planned personalized-investing project with RiskX, which would combine investment workflows with checks on agent authority, transaction intent and risk.

Verifiable records also feature in AEREDIUM’s October 21 explanation of its blockchain infrastructure. The company says independent verifiers can check its ledger without being able to alter it. That post concerns the underlying network; the new agent products bring their own signed decision records to coding and financial workflows.

𝕏Post from @aerediumView the original post on X↗

For businesses adopting agents, the value of these controls is practical: routine actions can proceed within an agreed mandate, exceptions can reach a person, and decisions leave a record. As agents gain access to both software and money, defining that mandate becomes part of deploying them.