Brussels/London — In a move that could reshape electronic foreign‑exchange markets across Europe, the European Securities and Markets Authority (ESMA) and the European Central Bank (ECB) published joint draft guidance in early August 2026 that specifically targets the growing use of artificial intelligence in algorithmic FX execution.

What the guidance requires

The consultative document, issued jointly by ESMA and the ECB, focuses on three interlocking requirements for dealers, multilateral trading facilities (MTFs), and third‑party algo providers operating in the European FX trading ecosystem:

  • AI explainability and logging: Firms must retain auditable decision logs for AI and machine‑learning components that influence execution, including model inputs, key intermediate signals and final routing decisions, for a minimum of two years.
  • Pre‑trade stress and slippage testing: Algo vendors and liquidity providers must run standardized pre‑trade stress tests against adverse‑market scenarios (flash crashes, extreme spread widening, and liquidity withdrawal) and publish aggregated slippage statistics and worst‑case simulated outcomes for institutional counterparties.
  • Model governance and third‑party oversight: Firms must document model risk management frameworks, validation routines and vendor oversight processes, and must notify counterparties when material changes to an execution model are deployed.

Regulators said these measures are designed to reduce operational risk, improve market resilience and increase transparency around automated decisioning that can propagate liquidity shocks through FX venues.

Why FX, why now?

The guidance reflects two converging trends: the rapid proliferation of machine‑learning‑driven execution algos across spot and NDF (non‑deliverable forward) venues, and heightened regulatory scrutiny of model risk and AI systems across financial markets.

Since 2024, many FX desks and independent execution providers have progressively embedded reinforcement‑learning and other ML techniques to optimize routing and microstructure decisions. While those tools can reduce trading costs and improve fills in normal markets, regulators warn they may also behave unpredictably in stressed conditions or amplify liquidity withdrawal when multiple participants rely on similar model logic.

Market reaction: compliance scramble

Market participants reacted quickly. Major bank liquidity providers and top independent algo vendors told reporters they expect a significant one‑time compliance cost related to data retention, audit tooling and expanded testing workflows.

Smaller ECN operators and boutique algo shops said the requirement to publish standardized slippage statistics will be operationally intensive. “You’re not just changing a reporting screen — you need robust simulation infrastructure and a formal validation protocol,” said a head of execution at a midsize brokerage. “That will push some participants to partner with cloud vendors or withdraw products that aren’t economically viable to maintain.”

Execution costs and liquidity implications

Traders interviewed for this report warned that some market makers could raise fees or reduce displayed liquidity to offset compliance costs and higher model‑risk capital. Conversely, the guidance could benefit larger venues and consolidation‑minded cloud providers that can amortize tooling across many clients.

Institutional buy‑side desks welcomed the transparency measures. Portfolio managers and FX risk teams highlighted the value of standardized slippage analytics and explainability when comparing execution providers. “If you’re choosing an algo, you want to know how it behaves in a euro‑flash or when the USD funding market tightens,” said a global macro trader. “Structured disclosure removes guesswork.”

Operational specifics and timeline

The draft sets a consultative period of 60 days and proposes a phased implementation: initial disclosure and logging requirements within nine months of final guidance, and full compliance for pre‑trade stress testing and published aggregated metrics within 18 months.

ESMA and the ECB also propose collaborative supervisory testing — coordinated “war game” exercises with selected market participants to validate test scenarios and assess systemic impact before full implementation.

Key questions and potential frictions

  • Intellectual property and model secrecy: Firms argue that explainability obligations must be balanced with IP protection. Regulators have proposed allowing aggregated or redacted disclosures that preserve commercially sensitive elements while still providing meaningful insight.
  • Cross‑border enforcement: Many FX counterparties operate globally; the guidance urges national competent authorities to coordinate to avoid conflicting requirements, but implementation across jurisdictions could be uneven.
  • Data standardization: Traders want regulators to endorse specific formats for slippage metrics and stress scenarios to make provider comparisons practical.

Broader industry implications

If finalized, the guidance would likely accelerate several trends already underway in FX markets:

  1. Consolidation among smaller algo vendors that lack scale to build compliant testing frameworks.
  2. Greater reliance on third‑party validation and independent model auditors.
  3. New market‑data products that package standardized slippage and stress reports for buy‑side procurement teams.

For traders, the short‑term outlook is mixed: potential for slightly wider spreads and reduced displayed depth as providers recalibrate risk, balanced by better information to select execution partners and lower tail‑risk from opaque AI decisioning.

Next steps

Market participants have until early October 2026 to submit responses to the consultation. Industry groups — including global sell‑side trade associations and buy‑side execution committees — are already preparing coordinated feedback requesting clear templates, phased exemptions for smaller firms, and alignment with international AI oversight frameworks.

The final guidance will be closely watched beyond Europe. Regulators in North America and Asia have signalled interest in similar AI transparency measures, and any European rule could set de‑facto global standards for algorithmic execution in FX markets.

For FX traders, the immediate priority is operational readiness: begin inventorying execution models that incorporate AI, review logging capabilities, and engage vendors on proposed slippage metrics. Firms that move early are likely to gain a competitive advantage as counterparties increasingly demand documented, testable execution performance.