{"text":[[{"start":5.82,"text":"The writer is chair of BBVA"}],[{"start":8.04,"text":"Millions of people are starting to discover what personal AI agents can do for them. These assistants make bookings, review and pay bills, file insurance claims, and carry out endless other time-consuming or complex tasks. In finance, they promise to give everyone the support and personalised advice that only the wealthy could afford until now."}],[{"start":27.66,"text":"Companies, unfortunately, are not yet ready to deal with personal AI agents. Few have a dedicated channel for the agents to access their services. This means agents have no choice but to come in through the back door, “pretending” to be their users by logging in with their passwords."}],[{"start":44.32,"text":"Companies can spot some of these agents, and many block them, but others get through and are then free to operate as the real user would. This exposes the user to various risks, from compromised credentials — it’s never a good idea to share your passwords — to unauthorised access or rogue agent behaviour. Companies, in turn, may face growing disputes from customers refusing to accept actions their agents might have taken."}],[{"start":68.34,"text":"Keeping agents out to prevent these risks is hardly the answer. While many companies, including BBVA, will offer their own AI agents, customers must also be free to work with any agent they choose, and to decide which third-party services (and corresponding data) their agents may access. The way forward is for agents to gain this access with their own “agent identity”, not impersonating the user."}],[{"start":90.8,"text":"In finance, some of this is already law. Under Europe’s payment rules, firms that access customers’ accounts or make payments for them must be registered or licensed and identify themselves to the bank, which in turn must give them a secure interface. Since 2019, Europe’s banking regulator has made clear that these firms may no longer get in by pretending to be the customer. Agents that provide these services are no exception."}],[{"start":115.42,"text":"But these rules cover only part of what agents do. Every sector, finance included, needs a “Know Your Agent” framework: any service should know which agent it is dealing with, who is accountable for it and what it may do."}],[{"start":127.74,"text":"Customers should sign in with the service provider itself and grant permissions there. Those permissions must be very specific: an agent may check balances and pay utility bills, but not add a new payee or take out a loan. They must be revocable at any time, and every action should leave a record."}],[{"start":146.12,"text":"The industry is already building the standards. BBVA was the first European bank to join the Personal Agent Protocol, an initiative also known as “poppy” that is led by enterprise AI start-up Sierra and Meta. Creating one shared standard is far better than each company inventing its own, but regulatory action is still needed urgently."}],[{"start":167.08,"text":"What needs to be done? Europe’s AI Act requires AI systems to disclose themselves to people. It should also require agents to identify themselves to the services they access and, after a transition period, ban impersonation. In return, large service providers, from banks and utilities to online platforms and social networks, should be required to offer identified agents a standard channel, on fair terms."}],[{"start":191.14,"text":"We also need a system that identifies legitimate agents, rather than allowing each company to decide which ones it can trust. Every agent should carry a verifiable credential, building on eIDAS, Europe’s cross-sector framework for digital identity. Where an agent carries out regulated activities, its provider should also hold the corresponding licence."}],[{"start":211.18,"text":"Finally, there needs to be a clear allocation of liability, which should follow control. Customers should be responsible for the permissions they grant and the instructions they provide to the agents. Service providers should be liable for actions taken beyond those permissions and agent providers for actions taken beyond those instructions. And all providers should bear responsibility when their own systems are breached. Customers should know in advance where to turn to resolve the problem and should not bear losses beyond their control."}],[{"start":239.98,"text":"These rules and standards can and should be applied across sectors, from the outset, to adequately protect consumers wherever agents work for them. In the same vein, common rules and standards should operate across borders, so that an agent trusted in one jurisdiction can be trusted in another."}],[{"start":256.34,"text":"The early success of personal AI agents suggests they are here to stay. Instead of blocking them or letting them operate in the shadows, we should open the front door."}],[{"start":265.7,"text":""}]],"url":"https://audio.ftcn.net.cn/album/a_1794583029_5275.mp3"}