Technical deep-dives, GDPR guides, and development updates from the euRedact team.
Most redaction removes identifiers: names, ID numbers, emails. The GDPR's strictest rules are for something else: health data and the other special categories under Article 9, and criminal-offence data under Article 10. None of it has a format you can match. Here's why it matters, why it's hard, and how covering it lets sensitive internal processes use AI with far less risk.
The managed neural layer we've been building alongside the open-source engine is running in closed alpha. Structured PII never reaches the model, inference stays in the EU, and zero data retention is on by default, so your document text is gone the moment your result is handed back. Here's what it does — and what still has to be finished before anyone can sign a contract for it.
This year, two of Europe's most security-conscious institutions were hit by breaches — one through a zero-day no patch existed for. The lesson isn't that anyone failed at security. It's that the size of a leak is decided years before the attacker shows up, by how much data you keep. Data minimisation is attack surface reduction.
The second-generation euRedact redaction model is through its first full evaluation run. Raw potential misses per document dropped 54% versus the first generation, with the biggest gains on exactly the entity types that matter most under GDPR. Here's what the numbers show, what changed under the hood, and why the headline benchmark is still a few weeks out.
Two weeks before the AI Act's biggest compliance deadline, Brussels pushed high-risk obligations back by 16 months. Here's what actually changed, what didn't — and why reading this as 'compliance can wait' is the most expensive mistake you could make with it.
Log files are one of the most persistent and underappreciated sources of PII leakage in production systems. Here's what ends up in them, how it gets there, and what you can do about it.
A new POLITICO survey shows overwhelming European distrust of US and Chinese tech companies handling personal data. For companies building data-intensive products in Europe, this isn't just a PR problem — it's an architecture problem.
US-built PII tools miss most European identifiers. Here's why redacting PII across 31 EU countries requires a fundamentally different approach — and what we're building to solve it.
We've trained our first dedicated PII detection model for European documents. It's faster and cheaper than general-purpose LLMs, and the early evaluation results are promising.
New posts go out over RSS. No address to hand over, no list for us to keep — which felt like the right default for a site that argues for data minimisation.