Security and Recovery in the Agentic Economy | CoinCover, Fireblocks | FF Virtual Arena #370

Quick Summary

In this FF Virtual Arena, Anthony Yeung of CoinCover, Matthew Foyle of Fireblocks, and Josh Goodbody of BasedAI join Ali Paterson to discuss the rise of the agentic economy. As AI agents transition from providing information to executing financial transactions, the panel explores the infrastructure required to manage security, control, and asset recovery.

What is the current state of the agentic economy?

The shift from agents that simply answer questions to agents that perform real world work is already underway. Josh Goodbody of BasedAI notes that agents are beginning to buy products, pay for services, and conduct payments with one another. While current transaction volumes remain relatively low, with approximately 15 to 20 million dollars moved this year, the trajectory is significant. McKinsey estimates that AI agents could move as much as 3 trillion dollars by 2030. This evolution opens a Pandora's box of questions regarding responsibility and the need for robust guardrails. Anthony Yeung of CoinCover highlights that there are two primary flavours of these payments: machine to machine transactions and agentic commerce using traditional fiat rails. While fiat transactions benefit from established precedents set by major card networks, the machine to machine world is still establishing its infrastructure.

What are the primary security and recovery challenges?

When AI agents hold keys and move money at machine speed, new failure modes emerge. Anthony Yeung of CoinCover expresses concern over the lack of independent recovery mechanisms. While individual transactions might be small, the aggregate value across thousands of agents can become substantial, attracting the attention of regulators and auditors. The challenge lies in ensuring that funds can be recovered if an agent fails or a key is lost. Matthew Foyle of Fireblocks explains that developers face critical architectural decisions, such as whether an agent should have its own identity or inherit one from a human. There is also the question of where to place authority: at the credential level or at the signing layer through a policy engine. Without these technical guardrails, the industry may struggle to satisfy regulatory requirements for safety and remittance.

How is liability determined when an agent makes a mistake?

The question of who is on the hook for an autonomous transaction remains a significant hurdle for institutional adoption. Josh Goodbody of BasedAI suggests that, in most jurisdictions, the agent owner will likely be held responsible for the actions of the entity they deployed and funded. However, chained transactions involving multiple intermediaries could complicate legal frameworks. Major financial institutions like JP Morgan and Bank of America have indicated they will not grant agents unilateral access to banking rails without clear human accountability. Anthony Yeung of CoinCover notes that traditional players like Visa and Mastercard will play a vital role in building trust through mediation and dispute resolution, similar to their role in modern e-commerce.

What must happen for institutions to adopt agentic payments at scale?

For large scale deployment to occur within the next 18 months, the panel agrees that several conditions must be met:

• Clear governance frameworks that provide end to end observability and accountability.

• The presence of a human on the hook to ensure every decision has a defined owner.

• Advanced policy engines that allow organisations to set risk appetites and control agent actions.

Matthew Foyle of Fireblocks emphasises that institutions need enough data and visibility to make informed decisions about their risk exposure. Anthony Yeung of CoinCover adds that the industry must provide the technology for these controls today, even as regulation continues to evolve in parallel. The recent demo at Sibos of automated interest earning portals illustrates the potential for disruption, but also the need for stability.

FF NEWS TAKE:

The transition to an agentic economy represents a fundamental shift in how value moves across the internet. While the potential for efficiency gains is enormous, the industry is currently in a race to build the safety nets that regulators demand. The focus on recovery and policy engines suggests that the future of AI commerce will not be entirely autonomous, but rather a hybrid model where machine speed is tempered by human oversight and institutional grade security.