Safety & Policy
10 articlesThe EU AI Act and the People Who Will Have to Enforce It
Writing a technology law is the easy part. Staffing national supervisors, judging models from documentation, and defining high risk is the hard part.
Responsible Scaling Policies, Explained for Normal People
A lab publishes a table of capability levels and promises not to cross one without safeguards. The thresholds, evaluations and verdict are all its own.
What Congress Actually Learned From a Year of AI Testimony
Executive appearances made headlines, but the legislative output was smaller. What a year of US AI hearings actually produced, and who never testified.
The AI Safety Institute: A New Kind of Public Body
Too technical to be a regulator, too official to be a research lab. What the London and Washington institutes can build, and what they cannot.
Open Letters and Pledges: A Year of Safety Theatre
A pledge to pause development of something nobody is close to building costs nothing. Four questions that separate a pledge from a sentiment.
Why AI Watermarks Keep Failing the People Who Need Them
Provenance labelling survives compression but dies at a screenshot. Detection, metadata and registries all fail differently, and false positives may be worst.
Model Evaluations and the Messy Business of Measuring Danger
Before a model can be called safe, someone has to define what safe means. Elicitation, baselines and generalisation are where that argument starts.