Banks face *live AI audits*, Swedish cybersecurity rules start next week
Banking regulators want proof now, Sweden's cybersecurity law takes effect in days, and a medical AI case shows what can go wrong.
- 01
Banks asked to prove AI agents during exams
Banking regulators are now asking banks during examinations to produce decision logs, approval records and governance documentation for AI agents. Banks that can't demonstrate auditability face immediate supervisory consequences.
Why it matters for evidence: Article 12 requires automatic logging. DORA requires logs protected against tampering. Together they mean: keep logs, and prove they hold up.
Source: aigovernance.com
- 02
Sweden's Cybersecurity Act starts 15 January
Sweden's Cybersecurity Act enters into force on 15 January 2026. From 1 July 2026, essential and important entities must report incidents within 24 hours and file a detailed report within 72 hours.
Why it matters for evidence: The law wants reports within 24 hours. A sealed timeline of when you knew, and what you did, makes that straightforward.
Source: instagram.com (in Swedish)
- 03
Most enterprises unprepared for AI Act high-risk rules
Three-quarters of large enterprises lack preparation for EU AI Act high-risk obligations, and 74% have no compliance owner. Four in five AI deployments have no named accountable party.
Why it matters for evidence: Article 26 says users must monitor AI, keep logs for six months and explain decisions. Without records, those duties are just words.
Source: techtimes.com
- 04
Stigmatising language in records skewed AI triage
A study shows stigmatising language in medical records can cause AI triage systems to deprioritise high-risk patients in emergency departments. The case raises whether providers can prove what the AI decided and why.
Source: meditude.se (in Swedish)
- 05
Technical guide to trustworthy AI audit trails
A technical guide explains how to build audit trails for AI agents using correlation IDs, tamper-evidence, clock synchronisation and retention policies aligned with NIST controls.
Source: moring.ai
- 06
Workday AI hiring case proceeds to trial
A US federal judge has allowed a class-action lawsuit to proceed against Workday, alleging its AI hiring tools discriminate against applicants over 40. The case highlights vendor liability and the challenge of proving what AI systems did.
Source: facebook.com