AEREDIUM Launched Guardrails for Agents That Move Money and Write Code: Governance Becomes a Product | inicIA
Direct answer
AEREDIUM, an Austin-based digital-asset infrastructure firm, announced on October 22, 2026 (via Chainwire) two products for governing autonomous AI agents: The Chair (for agents that write code) and MCP Connector (for agents that pay and trade). Both —launched Oct 13 on its new aeredium.ai platform— place a policy engine between the agent and the system it wants to touch: every proposed action is evaluated against the owner's rules and gets one of three outcomes (approved, denied, or held for human review), with each decision signed and recorded for later audit. Records are kept for 7 years. The signal: agent governance stops being an internal document and becomes software you buy.
What Each Product Does
- The Chair: governs AI coding agents (integrates with Claude Code and OpenAI agents). Every proposed code change is approved, denied or held for a human before it runs. 30-day free trial available.
- MCP Connector: governs agents that move money, with two separate roles:
- Payer: the agent can send USDC, USDT, PYUSD and EURC only to owner-approved addresses, with no permission to trade.
- Trader: the agent can trade on designated exchanges inside its own wallet, with no permission to send funds to external recipients.
- Security is cryptographic: the wallet uses threshold signatures generated inside secure enclaves —no complete private key is ever created, and AEREDIUM holds no key to the owner's funds. Supports Ethereum, Arbitrum and Base.
- Designed for Claude, open to OpenAI: initial compatibility is with Anthropic's ecosystem, with openness to other frameworks.
Why This Is the Missing Piece for Scaling Agents
- It separates permission from capability: the classic mistake is giving an agent "access to everything" because it's easier. The right pattern —which these products package— is: the agent proposes, the policy decides, the human steps in only for the sensitive stuff.
- It turns audit into a byproduct: every signed, 7-year-recorded decision means compliance stops being a quarterly project and becomes a queryable log. In regulated industries, that's what separates a pilot from a real deployment.
- Roles matter more than permissions: making "paying" and "trading" distinct roles (Payer vs Trader) replicates in software the separation of duties finance has demanded from humans for decades.
The Line That Sums It Up
"AI agents are no longer just answering questions. They're writing code, moving money and making decisions that have real consequences. 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 of AEREDIUM.
The Honest Note
- It's the company's own announcement (Chainwire): capabilities are described by the seller. Before buying, demand a trial on your own use cases (The Chair's 30 days work for that) and independent security reviews of the design.
- The category is forming: 1Password launched its credential layer the same Oct 22 and Mural its payments infrastructure on Oct 21. Agent governance now has three layers being bought separately: identity, money and policies. Your architecture should account for all three.
Related links
- OpenAI disclosed 4 cases of agents crossing lines: the containment checklist for your business
- What to do when your AI agent makes a mistake
- How to tell if your AI is working: the 5 KPIs that matter
Frequently asked questions
What did AEREDIUM launch?
Two products announced Oct 22, 2026 (launched Oct 13): The Chair, which governs AI agents that write code, and MCP Connector, which governs agents that pay and trade stablecoins.
How do the controls work?
A policy engine evaluates each action the agent proposes before it runs: it approves it, denies it, or holds it for human review. Every decision is signed and recorded for 7 years.
Which coins and networks are supported?
USDC, USDT, PYUSD and EURC on Ethereum, Arbitrum and Base, with threshold signatures in secure enclaves.
Does this replace human oversight?
No: it turns oversight into the exception instead of the routine. The human sets the rules and reviews only what the policy holds. Full control stays with the owner, who can pause or delete the agent at any time.
