AI Pulse Daily Brief logo

AI Pulse Daily Brief

Archives
April 28, 2026

AI Pulse Daily Brief | 2026-04-28

AI Pulse Daily Brief | 28 April 2026

Reading time ~3 mins

Dutch cabinet opens AI Act implementation consultation with AP as lead supervisor and June 1 deadline. Unpatched MCP protocol flaw exposes banking AI infrastructure. Treasury and Fed brief bank CEOs on Mythos zero-day threat. MATCH Act advances ASML DUV ban toward floor vote. FCA launches live AI testing with Barclays and Lloyds.

Top signal

Dutch cabinet publishes AI Act implementation bill; AP designated lead supervisor, consultation closes June 1. Authority

Signal: Staatssecretaris Aerdts submitted the Uitvoeringswet AI-verordening on April 20, designating the Autoriteit Persoonsgegevens as the national competent authority for AI Act domains not covered by DNB or AFM, with the Rijksinspectie Digitale Infrastructuur in a coordinating role; consultation closes June 1.

Relevance: This resolves the central supervisory uncertainty for Dutch banks under the AI Act, and the June 1 deadline is a narrow window to shape national implementing rules before they enter parliamentary debate.

Consider: Assess whether to submit a position paper to the internetconsultatie before June 1, particularly on how AP's high-risk AI system oversight intersects with DNB prudential supervision.

Rijksoverheid

Security

Unpatched MCP flaw exposes up to 200,000 AI instances as Anthropic declines to remediate. Media

Signal: OX Security disclosed an RCE vulnerability in Anthropic's MCP SDK affecting an estimated 200,000 instances across 150 million downloads; Anthropic called the behaviour "expected" and declined to patch.

Relevance: Banks using MCP-compatible agentic tooling carry DORA third-party risk liability regardless of upstream remediation, and the vendor's refusal to fix makes this a design-level exposure for the AI vendor risk register.

Consider: Confirm which internal agentic AI deployments use MCP-connected tooling and verify input sanitisation at every tool-invocation boundary.

American Banker | Infosecurity Magazine

Treasury and Fed brief bank CEOs on Mythos zero-day risk; fewer than 1% of flagged vulnerabilities patched. Advisory

Signal: Sullivan & Cromwell documented an April 15 meeting where Bessent and Powell warned bank CEOs that Claude Mythos autonomously discovers zero-days across operating systems and browsers; fewer than 1% of vulnerabilities shared with vendors have been patched.

Relevance: An unprecedented government-to-bank-CEO briefing on a single AI model's threat signals that European supervisors are likely to follow within 12 months.

Consider: Assess whether the bank's patch management cadence is calibrated for AI-discovered zero-days at scale, before DNB or ECB issue analogous advisories.

Sullivan & Cromwell

Regulatory

Digital Omnibus trilogue today; GDPR Article 4 "relative approach" to personal data on the agenda. Institute

Signal: The April 28 trilogue includes a proposal under which whether data is personal depends on the means available to each controller, plus replacement of 27 national DPIA lists with one EU-wide template.

Relevance: For a bank with extensive data-linking capabilities, the relative approach could reclassify datasets currently treated as pseudonymised, requiring changes across data-sharing agreements and processing registers.

Consider: Check whether pseudonymisation classifications in the bank's data governance are absolute or controller-relative, since a political agreement today would signal the change is coming.

iubenda

Perspectives

The Mythos meeting focused on the wrong AI risk: fraud, not cyber intrusion, is the unaddressed threat. Corporate

Signal: Dr. Shlomit Wagman argued in Fortune that the Treasury/Fed Mythos briefing concentrated on cyber-intrusion while ignoring AI-generated phishing, voice clones, and synthetic video that convince customers to authorise transactions at near-zero marginal cost.

Relevance: The higher-probability threat to a retail bank is machine-generated social engineering that bypasses perimeter defences entirely, a risk that received no equivalent mobilisation in the government response.

Consider: Evaluate whether fraud controls are calibrated for machine-generated social engineering at scale or still built for pre-generative-AI fraud volumes.

Fortune

Netherlands & Sovereignty

MATCH Act clears House committee; ASML DUV ban advances with 150-day Dutch response window. Media

Signal: The MATCH Act passed the House Foreign Affairs Committee on April 22, targeting ASML DUV lithography exports to China; Bank of America estimates a full ban could reduce ASML revenues by 14-15%.

Relevance: Committee clearance raises the probability of a floor vote and triggers a 150-day window in which the Dutch government must demonstrate equivalent controls or face unilateral US action via the Foreign Direct Product Rule.

Consider: Review whether investment and credit exposure to ASML-linked entities accounts for this revenue disruption scenario, with the Dutch response deadline around late September.

TechWire Asia

Industry & competition

UK FCA selects Barclays, Lloyds, UBS and five others for live AI testing with real customers. Media

Signal: Eight firms will test AI in live market conditions under FCA oversight from April through December 2026, with an assessment report due Q1 2027.

Relevance: This is the first regulator-supervised live AI testing programme in financial services, and European regulators have historically adopted FCA innovation approaches within 12 to 18 months.

Consider: Track Q1 2027 results as a leading indicator of what structured AI deployment pathways European regulators could introduce for Dutch banks.

Retail Banker International

On the radar

  • A financial services firm replaced its QA team with AI, saving $1.2M annually; the system then generated a pricing error costing $6M, a 5x loss-to-saving ratio. QA Financial
  • Anthropic acknowledged three engineering changes degraded Claude for up to six weeks without notifying enterprise users, including a reasoning downgrade and a caching bug. The Register
  • France is migrating its Health Data Hub from Microsoft Azure to Scaleway on CLOUD Act grounds, establishing the most concrete EU precedent for sovereign data repatriation. Euronews

Don't miss what's next. Subscribe to AI Pulse Daily Brief:
Powered by Buttondown, the easiest way to start and grow your newsletter.