When AI agents hold the keys: security, control and recovery in the agentic economy

An on-demand panel with Fintech Finance News on AI agent wallets, agentic payments and crypto recovery.

Who's on the hook when an AI agent makes a bad digital transaction
How to give agents autonomy without letting them drain a wallet
Why recovery infrastructure has to be built in from day one for agentic payments

Featuring

Anthony Yeung Speaker

CCO, CoinCover

Matthew Foyle Speaker

DevRel Lead, Fireblocks

Josh Goodbody Speaker

Co-founder & COO, BasedAI

Building with agents? Talk to us about recovery

What you'll learn about AI agent security and recovery

AI agents are starting to hold wallets and move money on their own. Crypto is final, and agents move at machine speed. Josh Goodbody (BasedAI), Matthew Foyle (Fireblocks) and Anthony Yeung (CoinCover) work through what that means for anyone building or running wallets.

Where are we really? Josh helped build Warden, one of the early agentic wallets. What's live today, and what's still a demo.
The new ways it goes wrong. Prompt injection, hijacked agents and agents confidently getting it wrong.
Autonomy vs control. Spending limits, pre-approved rules and agent identity. Where's the line between useful and dangerous?
When it goes wrong anyway. There are no chargebacks in crypto. Why recovery needs to be built in from day one rather than added after the first big loss.
Who's on the hook? The user, the agent developer, the wallet provider or the model provider. Josh brings the legal and compliance view from Huobi and J.P. Morgan.
What unlocks adoption? What has to be true in the next 12 to 18 months before institutions let agents touch real money at scale, plus one thing every builder should do now.

Want the background first? Read why agentic payments need independent crypto wallet recovery.

Watch the webinar

FAQs: AI agent wallets

What is an agentic wallet?

An agentic wallet is a crypto wallet that an AI agent can use to hold funds and authorise transactions on behalf of a person or business. Many are embedded wallets that let agents make frequent, low-value payments, often in stablecoins, for things like data, content, computing capacity and APIs.

What new risks do AI agents bring?

On top of familiar risks such as lost keys, hacks and human error, agents bring new failure modes: prompt injection, hijacked agents and agents confidently making the wrong transaction. There is also infrastructure risk. If an embedded wallet provider has an outage, loses signing capability or has key material compromised, every agent relying on it can be unable to transact at once.

Are spending limits enough?

They are essential, but they solve a different problem. Spending limits, approved counterparties and human approval steps reduce the risk of unauthorised spending. They do not restore access if the wallet infrastructure itself becomes unavailable. That needs a separate recovery route.

Why build in recovery from day one?

Crypto transactions are final, there are no chargebacks, and agents act at machine speed. Some forms of lost access become permanent if no alternative route was set up in advance. Recovery should be independent of the primary wallet infrastructure, governed by clear approvals and identity checks, and tested regularly, before agents are given control of funds.

Who is liable when an agent gets it wrong?

There is no settled answer yet. Responsibility could sit with the user, the agent developer, the wallet provider or the model provider, depending on how the system is designed and what has been agreed. The panel works through the legal and compliance angles and where protection fits in.

How does CoinCover help?

CoinCover provides independent wallet recovery that works alongside existing wallet infrastructure, giving businesses a tested route back to their digital assets if access is lost. Talk to us about protecting agent wallets.