OpenAI will watermark AI text in EU

OpenAI says it will roll out AI text watermarking by default in the EU to comply with Article 50. The watermark is machine-readable, so automated tools can detect AI-generated text.

Why it matters for evidence: Article 50 asks who provided the tool to mark the output. When someone later claims they never said it, the watermark is your starting point, not your answer.

Source: digital.nemko.com

AI Office opens first safety investigation

The Commission's AI Office is investigating OpenAI and other labs over safety incidents. The question is whether they can show compliance with safety rules and document what their systems did.

Why it matters for evidence: High-risk systems must log events automatically. If you can't show what happened during an incident, the investigation starts with a blank page.

Source: linkedin.com

Arize and Evroc demo AI audit trails

Engineering teams can now watch a demo of building observable AI systems with audit trails for the AI Act. It covers practical instrumentation for logging and tracing agent behaviour.

Source: youtube.com

AI medical device certifier raises $50m

Scarlet, a certification body for AI and software medical devices, has raised $50 million in Series B to accelerate device certification. Manufacturers need to prove their AI medical devices meet regulatory requirements.

Source: Tech.eu

IMY examines web scraping and data protection

Sweden's data protection authority is looking at whether collecting personal data through web scraping can meet good data protection practice. The challenge is proving the data is handled correctly and doesn't end up in the wrong hands.

Source: IMY: nyheter (in Swedish)

Machinery Regulation lands in January 2027

The EU's Machinery Regulation takes effect on 20 January 2027, and the remaining CRA obligations follow on 11 December 2027. Manufacturers of machinery, automation and robotics selling into Europe need to prepare for both.

Source: facebook.com