St Georges Strategy

Weekly brief / 27 Sept 2026

The dependency is now the decision.

Three immediate developments and two useful prompts point to one leadership question: when an external model, provider, market utility or component shapes an outcome, can its controls be seen, challenged and paused?

Five-minute read / one-minute scan

Weekly Judgement

Week ending 27 Sept 2026 · Edition 15

The dependency is now the decision.

Archive record · The Weekly Judgement published with this edition is retained here in full.

What happened

Anthropic has published a low-stakes book-trading experiment about agents acting on users’ behalf; ASIC has strengthened AI trading safeguards; ENISA has examined how dependencies weaken resilience; the Bank of England has consulted on its fees for existing payment-system supervision; and CISA has added newly exploited vulnerabilities.

Why it matters

The boundary is not the contract or platform label. It is whether management can identify the dependency, understand the permission or safeguard it affects, and act before a weak external link becomes a customer, conduct or market-integrity event.

What to do

Pick one decision-critical journey this week. Trace it from the outcome to every external model, data feed, provider and exploitable component; put the one-page chain beside the escalation call tree and test who can halt it.

The weekly readout

The dependency chain is the control.

The useful leadership question is not whether a supplier or model was approved. It is whether the dependencies that make a live outcome possible can be seen, challenged and paused.

01 / What happened

External services are moving into the decision path.

Anthropic has published a low-stakes book-trading experiment about agents acting on users’ behalf; ASIC has strengthened AI trading safeguards; the Bank of England has consulted on supervisory fees for existing payment-system oversight; ENISA has examined dependencies; and CISA has added newly exploited vulnerabilities.

02 / Why it matters

A contract does not show the live point of failure.

Permission, market safeguards, service-provider supervision and an exploitable component can all alter a customer or trading outcome. The live chain matters more than the label on the contract.

03 / What to do

Trace one journey from outcome to dependency.

Choose one customer or trading outcome. Follow it through every external model, data feed, provider and component that can change it.

04 / Evidence

Test the power to pause it.

Put the chain beside the escalation call tree. The useful evidence is who can see the exposure, intervene and decide before the external weakness becomes an incident.

Top 5

This week’s significant signals

Each title goes to the underlying publication. The Signals pages carry the wider source trail and retained material.

  1. 01

    Anthropic’s agent-trading experiment tests delegated-action controls

    Delegated action experiment
  2. 02

    ASIC makes AI trading safeguards part of market-integrity discipline

    Live trading safeguard
  3. 03

    Bank of England consults on fees for payment-system supervision

    Supervisory-fee consultation
  4. 04

    ENISA frames dependency concentration as a resilience exposure

    Service-chain resilience
  5. 05

    CISA's 25 September KEV update tests dependency containment

    Containment decision

Question

Which external dependency can change one decision-critical outcome — and who can halt it?

Start with a real customer or trading journey, not a supplier list.

Evidence

The service chain, live permissions, intervention authority, last test and current exposure

Ask for the facts that show the dependency can be seen and acted on.

Source trail

Anthropic, ASIC, Bank of England, ENISA and CISA

Each signal links to its primary source; the judgement is our editorial interpretation.

Regulatory Horizon

The dates behind the dependency decisions

Use the current regulatory horizon to see the official dates, evidence and action windows that sit behind the week’s operating questions.

Regulatory Horizon / updated 27 Sept 2026

Confirmed dates, current obligations and the evidence to prepare

A practical register of upcoming regulatory dates and their operating implications.

Reading the signals

Where the dependency sits

Three short read-across points for the next leadership conversation.

Delegated action

A low-stakes agent experiment still surfaces useful control questions.

Project Swap is not a financial-market deployment. Its value here is narrower: permissions, intervention and action records should be tested before an agent is trusted with a consequential decision.

Operating perimeter

The fees consultation applies to an existing supervisory perimeter.

Recognised payment systems and specified service providers are already within the Bank’s supervisory remit. The current consultation concerns the 2026/27 fees, not a new extension of that perimeter.

Containment

Dependency concentration and exploited components expose the same weakness.

A service-chain failure and a newly exploited flaw are manageable only when the affected dependencies can be identified, isolated and escalated in time.

Executive challenge

Three questions from the week

Carry one into a committee, 1:1 or control review.

  1. Which external dependency can change one decision-critical outcome — and who can halt it?
  2. Where can automation influence a trading or customer outcome without a tested override and a reconstructable record of why it acted?
  3. When an exploit or supplier problem appears, can we identify every affected dependency and decide on containment within one business day?