Kin lets Meta’s AI agent shop for insurance, Insurify blocks it. Who verifies the agent?

Meta’s Muse, Kin and Insurify · The FCA on open finance infrastructure for AI agents · The ACPR stops Wakam writing new business · Supervising open finance in Africa · Cancer survivors and fair access to financial services

Meta’s Muse goes insurance shopping: Kin opens the door, Insurify closes it

On 30 September 2026 Kin, a US direct-to-consumer home insurer, announced that Meta’s personal AI agent Muse is welcome to get quotes on its website. A week earlier, Insurify had blocked Muse from its comparison marketplace, arguing that a quote stripped of limits, deductibles and eligibility is not a fair comparison. Two insurance players, the same question and opposite answers: should an AI agent be allowed to shop for insurance on a customer’s behalf?

Kin says Muse got a home insurance quote with only a handful of questions, because Kin already holds most of the data about individual homes. That is the real lesson. The insurers that do well in agentic distribution will be the ones whose quote process is fast, data-rich and machine-readable, and Open Insurance is quickly turning from a policy discussion into a commercial necessity. In my blog post I look at why both companies are partly right, what it means for brokers and how insurance distribution could look in five years.

Read the full analysis on my blog.

The FCA asks what open finance infrastructure needs when AI agents act for customers

On 29 September 2026 the FCA’s Smart Data Accelerator published Powering open finance infrastructure, a research note summarising work it commissioned from Saïd Business School, University of Oxford, and Raidiam. It treats open finance as an infrastructure design challenge, not simply open banking extended to more products.

The Muse story above shows why. An agent arrives at an insurer’s website, and the questions are who verified it, who gave it the mandate and who is accountable when something goes wrong. The research puts the same questions at system level: where an AI-enabled service acts for a consumer or firm, the infrastructure may need to establish who operates the service, what authority it has and how that authority can be monitored or withdrawn.

It starts from the practical conditions that make data sharing usable, trustworthy and scalable: common technical standards, secure APIs, reliable accreditation and identity, effective consent and authorisation, operational monitoring and the ability to suspend or revoke access. It then compares three model families, each illustrated with an international example:

1. Centralised (UAE). A central trust framework, API hub and shared services. Strong central coordination reduces fragmentation but raises questions about concentration risk and scalability.
2. Decentralised or federated (EU, PSD2). The lesson drawn is that a legal right to data access does not guarantee consistent, interoperable implementation without stronger coordinating standards.
3. Hybrid (Australia and Brazil). Central accreditation and standards, with data exchange run by the participants. Australia shows that high-assurance accreditation can create an onboarding burden for smaller firms. Brazil expanded in phases, from open data on businesses to consumer data, payment initiation and finally investments and insurance.

The research concludes that “coordinated trust functions with distributed delivery” are a credible proposition for further UK testing: central coordination where consistency matters most, such as accreditation and trust, and flexibility in how firms build services. For agentic finance it points to OpenID Federation as one framework for sharing verified information about an agent’s identity, capabilities and authority across trust domains, and to crypto-agility so that keys and certificates can be updated as quantum risks grow. The next phase will test how authority is delegated to an AI-enabled service, interpreted across institutions and withdrawn by the consumer.

For EU readers, it is worth noticing that the EU appears as the decentralised example. The Financial Data Access Regulation (FiDA), still without agreement, would rely on data sharing schemes designed by market participants. Whether that model can carry agents acting across institutions is a question the UK is now testing first. Read the research note.

The ACPR bars Wakam from writing new business

On 25 September 2026 the French prudential supervisor, the Autorité de contrôle prudentiel et de résolution (ACPR), prohibited Wakam SA from underwriting new contracts and from renewing existing ones, and opened proceedings that could lead to the withdrawal of its authorisations. The measure is temporary, and existing contracts and claims continue to be handled.

I usually do not comment on individual companies, but this one is as much policy and innovation news as company news. Wakam is a white-label carrier: it provides the licence and the balance sheet, while MGAs, insurtechs and digital platforms design and sell the products. Many of Europe’s embedded and digital insurance models rely on carriers like this in the background. My blog post looks at what the case means for MGAs, for supervisors and for the wider innovation debate.

Read the full post on my blog.

Cenfri sets out how African regulators can supervise open finance once it is live

In September 2026 the Centre for Financial Regulation and Inclusion (Cenfri) published Oversight and supervision in open finance: Considerations for African regulators. Designing an open finance framework is only part of the task; regulators also need to know how the ecosystem will be overseen once it operates. The report covers three areas: operational and ecosystem oversight, conduct supervision and monitoring whether open finance actually delivers on inclusion and competition. Prudential supervision is out of scope, since open finance adds little new prudential risk to banks, while operational oversight and conduct are the first priorities.

The useful part is how priorities shift across four phases, each with its own key risk:

1. Building the foundations. Key risk: passive leadership, with no clear authority or industry buy-in. Supervisors make sure licensing, data access and supervisory powers are in place and set policy baselines.
2. Pilot. Key risk: a pilot that is not focused and lacks representative participants, so the system goes live with systemic problems. The focus is on operational testing and consent.
3. Scale. Key risk: participants do not comply and the consumer experience is poor, leading to too little data sharing. Supervision deepens on operations and conduct, including pricing and discriminatory treatment of participants.
4. Ecosystem expansion. Key risk: the benefits accrue to a small number of dominant participants and consumers. The focus moves to system-wide outcomes and to participants outside the lead supervisor’s perimeter.

The priorities stack rather than replace one another: what is set up in the foundation phase still matters at scale. The African context shapes the details. A few large institutions dominate many markets, pensions, insurance and investments are often split between several regulators, and consent journeys need to work on feature phones, for which there is limited global precedent. The phase 3 risk, benefits going mainly to the dominant players, is one that open banking regimes in Europe know well. The work was funded by the “Data Governance in Africa” Team Europe Action, which Estonia co-funds. Read the report.