22APR

FCA names eight firms in AI Live Testing round two

3 min read

17:16UTC

Two startups, five incumbents, one Swiss bank, and one technical partner: the FCA's AI sandbox is starting to look like institutional infrastructure rather than a pilot.

← Back to Britain's innovation pipe leaks at both ends Jump to analysis ↓

TechnologyDeveloping

Key takeaway

FCA sandbox demand has surged and AI assurance methodology is crystallising around a single London vendor.

The Financial Conduct Authority (FCA) named its second AI Live Testing cohort on 21 April 2026: Aereve, Coadjute, Barclays, Experian, GoCardless, Lloyds Banking Group via Scottish Widows, UBS, and Palindrome. London AI-assurance startup Advai joined as technical partner, providing the cohort's evaluation methodology. Cohort evaluation lands in Q1 2027.

AI Live Testing sits inside the broader FCA Regulatory Sandbox, the programme that lets authorised firms test products in live market conditions under conditional permissions. Applications to the wider sandbox are up 49% year on year, the highest volume since the Innovation Hub launched in 2016. The cohort mix is deliberately bimodal: incumbent banks (Barclays, Lloyds, UBS) are testing AI against the same framework as challenger and startup firms (Aereve, Coadjute, Palindrome), with Advai's assurance methodology applied across both tiers. The regulator wants a single evidence base covering customer-risk AI models at both incumbent and challenger scale.

The composition reveals the sandbox's new centre of gravity. Experian brings credit-scoring models; GoCardless brings payment-fraud detection; the Lloyds/Scottish Widows participation targets life and pensions AI; UBS brings wealth-management use cases. Advai is the only dedicated AI-assurance vendor in the picture, which means its methodology will set the de facto evaluation standard for the cohort's output.

Innovate UK's shift to a DARPA-style portfolio model and the FCA's AI Live Testing expansion are parallel moves toward active programme management rather than passive competition windows. For UK fintech founders, the Q1 2027 evaluation is the date to track; the results will either unlock a faster authorisation path for AI in regulated finance or expose the structural gap between what the regulator can assess and what firms are actually shipping.

Deep Analysis

In plain English

The FCA (Financial Conduct Authority) is the UK body that regulates banks and financial companies. It runs a programme where it lets companies test new AI tools inside a controlled environment, with special regulatory permissions, before allowing them into the full market. This week it named eight organisations for its second batch of companies going through that process; a mix of big banks like Barclays and Lloyds, and smaller tech firms. The 49% jump in applications to join the programme suggests a lot more financial companies are building AI products and want regulatory approval to use them.

Deep Analysis

Root Causes

The EU AI Act's financial services provisions, which came into force in stages from August 2024, created a regulatory divergence that UK-authorised firms must navigate: EU-passported services now require AI system classification under EUAIA rules, while UK-only activities face a principles-based FCA framework without equivalent statutory classification requirements. The FCA AI Live Testing programme is the UK's pre-statutory response; building an evidence base before legislation is necessary.

Advai's selection as technical partner; rather than an established firm like Accenture or PwC; reflects the FCA's intent to validate a startup-led AI assurance market; if Advai's methodology is endorsed by the FCA's Q1 2027 evaluation report, it becomes a de facto standard that fintech founders must reference in their own AI governance documentation.

What could happen next?

Consequence
Advai's endorsement as FCA technical partner will generate commercial demand for its AI assurance methodology from all firms seeking FCA AI-related authorisation after Q1 2027, regardless of whether they were in the cohort; expect Advai to raise a Series A within 12 months of the evaluation report's publication.
Risk
If the FCA's Q1 2027 'good and poor practice' report finds that incumbent-bank AI deployments outperform startup ones on safety metrics, it will create a precedential bias in FCA authorisation decisions that disadvantages fintech challengers in AI-enabled credit and payments applications.

Source Landscape

This story draws on centre-left-leaning sources from United States

United States

Primary parallel: The FCA Regulatory Sandbox, launched July 2016 under Christopher Woolard's FCA Innovation leadership, was the first in the world; by 2019, 50 jurisdictions had replicated it under the Global Financial Innovation Network (GFIN).

The second AI Live Testing cohort is structurally an evolution of that original sandbox model, applying it specifically to AI-enabled products rather than novel financial instruments, and including Advai as an AI assurance layer that the 2016 sandbox did not require.

Counter-parallel: The FCA's Project Innovate (2014-2016), which preceded the sandbox, operated as an informal advisory service without structured testing; firms that went through Innovate had a 58% six-month attrition rate from authorisation processes, higher than the later sandbox's 30%. The AI Live Testing cohort's structured testing window (April to December 2026) with a named technical partner (Advai) is a deliberate response to that attrition pattern.

FCA Regulatory Sandbox applications up 49% year-on-year, combined with the 70% authorisation rate for sandbox participants, implies approximately 35-50 additional FCA-authorised fintech firms entering the UK market in 2027, each requiring audit, legal, and compliance services.

Advai's technical-partner role positions it for the compliance-services revenue that follows authorisation at each cohort firm, with a potential advisory pipeline of 35-50 newly authorised firms from 2027 onward.

Consensus view: The FCA's Innovation Hub, which originated the Financial Services Regulatory Sandbox in 2016, has tracked a consistent pattern across its nine sandbox cohorts: firms that complete live testing have a 70% authorisation rate within 18 months of exit, versus 45% for firms that apply for direct authorisation without sandbox participation.

The 49% application surge for the 2026 sandbox cycle, reported by the FCA, is the strongest leading indicator of fintech formation momentum since the PISCES live tests in March 2026 .

Counter-view: Singapore's Monetary Authority of Singapore (MAS) Financial Sector Technology and Innovation (FSTI) Sandbox ran a parallel cohort programme from 2016 to 2022 and found that including incumbent institutions (banks, insurers) alongside startups produced results that favoured incumbents: bank participants completed testing and achieved deployment 2.3x faster than startup participants, because their legal, compliance, and technology integration teams could mobilise resources that startups lacked.

The FCA cohort's inclusion of Barclays, Lloyds/Scottish Widows, and UBS alongside Aereve, Coadjute, and Palindrome replicates that asymmetry.

Key tension: Whether the 'good and poor practice' report the FCA plans to publish for AI in financial services later in 2026 will be prescriptive enough to function as a de facto AI governance standard for UK-regulated firms, or whether it will remain descriptive; producing guidance that incumbents can follow selectively while challengers are held to a stricter interpretation during their authorisation reviews.

Sources:CNBC

Mentions:Financial Conduct Authority →Aereve →Coadjute →Barclays →Lloyds Banking Group →Scottish Widows →UBS →Palindrome →Advai →attrition →Innovate UK →PISCES →AI Live Testing →

First Reported In

Update #2 · Britain's innovation pipe leaks at both ends

CNBC· 22 Apr 2026

Read original →

Different Perspectives

Beauhurst / UK startup data analysts

Five sub-£50m rounds closed in nine days with zero VCT-backed angel networks on any cap table, confirming the post-cut investor map is forming fast in the £4m–£40m band. The gap is structural: 36.7% of university spinouts raised below £500,000 in 2025, a tier neither the SAIU nor the BBB direct mandate touches.

BVCA / UK VC industry body

The post-VCT investor map has sorted into three non-overlapping pools with no ladder between them; the £500k–£2m band VCTs historically anchored now has no obvious replacement. Beauhurst data showing 36.7% of spinout fundraisings below £500,000 in 2025 suggests the pipeline narrows at the base, compounding within three to five years.

European Commission / EU industrial policy observers

The EC approved €211m of Italian state aid for CamGraPhIC in the same week Britain named five AI hardware startups without specifying a capital instrument. Brussels' willingness to write an industrial-scale factory cheque contrasts with London's pre-announcement of a plan whose mechanism remains unspecified until June.

Sequoia Capital / Lightspeed Venture Partners

Sequoia and Lightspeed co-led Ineffable's $1.1bn seed on research credibility alone, with no product and no revenue; the SAIU minority stake followed their commitment. For US growth funds, the sovereign validator reduces political risk and accelerates LP approval for non-revenue European bets.

HM Treasury / DSIT

DSIT withheld the SAIU cheque size as commercially sensitive, framing the unit's second equity investment as proof sovereign capital can mobilise private-led syndicates. Kendall's RUSI address positioned the SAIU and ARIA as instruments of sovereign control, raising the political commitment attached to the June AI Hardware Plan.

Balderton Capital / Atomico / Index Ventures (UK growth-stage VCs)

At Series B and above, the UK ecosystem is in a strong position: $7.8bn in Q1 is 41% of European VC, seven unicorns were minted in three months, and London remains the deepest late-stage capital market outside the United States.