Courts, hospitals and *hiring tools*: evidence on trial
Four stories this morning, one theme: the harder it gets to trust digital records, the more it matters to seal them.
- 01
US courts wrestle with deepfake evidence
State courts are grappling with how to authenticate digital evidence as AI makes it trivial to alter images, audio and video. The challenge is proving what a system actually produced and whether the evidence is genuine.
Why it matters for evidence: Article 50 requires deepfakes to be labelled, but that doesn't answer the court's question: is this the original, or has someone changed it since?
Source: legalnews.com
- 02
California regulates AI in healthcare
Three new California laws require healthcare providers to exercise independent judgment when using AI clinical decision support, impose bias-mitigation and documentation duties on developers, and treat consumer health chatbots as providers under medical confidentiality law.
Why it matters for evidence: High-risk AI for triage and diagnosis will require automatic logging from December 2027 (Article 12), and proving what the AI recommended becomes part of compliance.
Source: hklaw.com
- 03
Hospital AI allegedly delayed intensive care
An AI-assisted bed-allocation system reportedly understated a patient's condition; the patient died hours after transfer. The incident raises whether the provider and hospital can prove what the AI system did and whether human oversight was in place.
Source: AI Incident Database
- 04
Swedish hiring tool combines AI with human control
Tengai AB offers automated screening combining structured AI-driven interviews with human control. The system falls under high-risk AI for recruitment, requiring logging, human oversight and the right to explanation from December 2027.
Why it matters for evidence: Article 12 requires automatic logging for high-risk recruitment AI, and Article 86 gives applicants the right to a clear explanation of the AI's role.
Source: mynewsdesk.com (in Swedish)
- 05
Every AI decision needs a name and a record
A governance framework argues every AI decision must have a named owner, undergo risk-proportionate review, and be documented in an auditable record for boards, auditors and regulators.
Source: dataproducts.io