Capability isn't the limiter for AI, credibility is

Big Picture

What happened: capability isn’t the limiter — credibility is. As AI gets cheaper and more embedded, the gating factor becomes trust: who is accountable, what gets tested, what regulators expect, and what remains uniquely human.

The Through-line

Across professions, payments, and even fashion, the winning playbook is converging: prove control, protect identity, and operationalise judgement.

1) Trust in AI surpasses accountants — and that should make the profession pause

What it is: Nick Best flags research suggesting people may trust ChatGPT more than human accountants in some scenarios — not because the model is “accountable,” but because it’s perceived as fast, objective, always available, and free.

Why it matters: “Trust without understanding” is the trap: models don’t carry professional responsibility, regulatory accountability, or commercial nuance unless prompted — and they don’t challenge risky decisions across the table. The profession’s value proposition is judgement, context, and ethics — not producing numbers.

First move: Build an AI-assisted advice checklist : what was machine-generated, what was verified, what assumptions were used, and who signs off. Make accountability visible.

Source: LinkedIn

2) Model upgrades don’t automatically improve outcomes — prompts do half the work

What it is: A study shared via MIT Sloan highlights that only about half of the performance gains from using a more advanced model come from the model itself; the other half comes from how users adapt prompts.

Why it matters: If your organisation “hot swaps” models in production, you risk silent regressions — and the problem compounds in multi-agent systems where small prompt shifts cascade into different tool actions.

First move: Treat upgrades like releases: maintain a prompt regression suite (golden prompts + expected behaviours), run A/B tests, and log where workflows break.

Source: LinkedIn

3) Practical AI governance: translate standards into executable controls

What it is: Victor Montefiore proposes a practical AI governance framework anchored on the NIST AI RMF , mapped back to the EU AI Act and ISO/IEC 42001, broken into 25 governance dimensions with lifecycle oversight (Define → Approve → Operate → Review).

Why it matters: Most firms can quote principles, but they struggle to implement them. The differentiator is operational design: ownership across the three lines of defence , traceability to requirements, and controls that actually run in the workflow.

First move: Start with an AI inventory/register and a simple RACI: business owns outcomes, IT owns technical control integrity, risk sets standards/challenge, audit tests independently.

Source: LinkedIn

4) Central Bank of Ireland: 2026 priorities raise the bar for payment & e-money firms

What it is: Finvisor’s digest of the CBI’s 2026 supervisory priorities centres on fraud and consumer protection , evolving AML expectations (including AMLA), deeper scrutiny of DORA readiness, embedding the revised Consumer Protection Code , and a sharper focus on financial resilience .

Why it matters: Supervision is shifting from “do you have a framework?” to “does it work under stress?” — especially around ICT risk, third-party dependency, incident handling, and whether firms can withstand funding shocks. It also calls out a harder fintech reality: if AI compresses cost advantages, incumbents may scale faster, changing the investment case.

First move: Run a readiness sprint: fraud outcome review, AML risk assessment refresh, DORA control testing, CPC process walk-through, stress-test funding , and update the wind-down plan .

Source: LinkedIn

5) AI investment in finance is moving at the speed of trust, not innovation

What it is: Bottomline argues B2B payments leaders are funding AI when it improves control — not novelty. Citing a report with a survey of 100+ UK/EU finance decision-makers , it notes 47% are deploying or piloting agentic AI, while 36% have no current plans (reflecting caution).

Why it matters: Use-case priority skews toward fraud detection, regulatory monitoring, and AML/KYC — “rules enforcement” over personalisation. Data sovereignty becomes strategic: 73% say it’s critical/important (41% “critical”), yet 64% report no formal sovereignty strategy.

First move: Build a “trust pack” for AI: sovereignty stance, dependency map, audit trails, explainability approach, and incident playbooks — then link investment decisions to reduced loss + stronger compliance defensibility.

Source: LinkedIn

6) AI in fashion and art: speed is the barrier — identity is the moat

What it is: Sahar Hashmi frames AI’s arrival in fashion and art as friction removal at scale — citing estimates of $6B in cost savings and ~20% EBIT expansion , plus consumer behaviour shifts (e.g. reported growth in generative-AI shopping).

Why it matters: When aesthetics can be replicated quickly, defensibility moves to brand architecture : narrative, cultural authority, and human resonance. AI expands what can be made — but can’t manufacture meaning.

First move: Use AI to kill production friction, but invest savings into what’s non-replicable: story, community, distinctive POV, and human craft.

Source: LinkedIn

7) “FinTech is dead, long live finance”: the curriculum signal

What it is: Daniel Liebau shares the path from a LinkedIn idea to a published opinion piece: FinTech should stop being a separate “course category” and be integrated into the finance curriculum — with a timeline from Sept 2025 post to March 2026 publication.

Why it matters: Industry doesn’t operate in silos; neither should education. If fintech is now simply “how finance works,” then skills need to be embedded: data, infrastructure, regulation, product, risk, and incentives — not treated as an elective.

First move: Replace “FinTech week” with “finance, rebuilt”: embed payments rails, AI governance, data sovereignty, and digital risk into core modules — assessed through real operating scenarios.

Source: LinkedIn