Fake experts, agent decisions and the evidence gap
Four stories today on what happens when you can't show what your AI did, and two on rules that are coming.
- 01
AI-generated fake experts sold products on YouTube
Fake testimonials and expert endorsements, created with AI, misled consumers in YouTube advertisements. The OECD AI Incidents Monitor logged the case.
Why it matters for evidence: Article 50(4) requires disclosure when AI creates or manipulates content that resembles real people and would pass as authentic.
Source: OECD AI Incidents Monitor
- 02
Europe's new product liability rules extend to software
The new Product Liability Directive makes manufacturers liable for defects after updates they control. It covers standalone software and AI, with evidence disclosure rules to help claimants.
Why it matters for evidence: Courts can order defendants to hand over relevant evidence, so being able to show which version ran when matters.
Source: legal500.com
- 03
EU's machinery regulation applies from 20 January 2027
The regulation replaces the machinery directive and requires safety-relevant software changes to be logged and traceable for five years. Extra rules apply to machines with self-learning behaviour.
Source: ventum-consulting.com
- 04
GDPR-compliant AI hosting guide published in German
A guide covers data processing agreements, data quality, logging and human oversight obligations for AI hosting. It addresses requirements that take effect before August 2026 under the Omnibus amendments.
Source: coding9.de (in German)
- 05
Less regulation may make clinical AI harder to sell
A healthcare AI expert argues that hospitals still require clear evidence of performance, failure modes and accountability. Regulatory frameworks provide the consistent evaluation standard that buyers depend on.
Source: linkedin.com