TUESDAY, MAY 26, 2026|No. 1
Technology · FI · US · FR

Anthropic, OpenAI, and Mistral sign "Helsinki Lines" on model evaluation

A shared evaluation harness — but with an explicit carve-out for open-source release. The decision is expected to draw a public counter-statement from the EleutherAI consortium within the week.

FI. A shared evaluation harness — but with an explicit carve-out for open-source release. The decision is expected to draw a public counter-statement from the EleutherAI consortium within the week.
23 sources
Pipeline ingest
3 reads
Positive / Neutral / Negative
3 countries
Related coverage

Three frontier labs have signed a shared evaluation harness with an open-source carve-out. The provisional text defines a shared threshold, establishes a new authority, and sets a 14-day clearance window. Three member states have confirmed the draft.

Trialogue is expected within the week. The regulation, if it survives, would enter force eighteen months after publication.

PAN's models place that probability — survival of the text in substantially its current form — at 51%, with the shared threshold the most likely point of compromise. Effects on trade volume cannot be modelled until enforcement-protocol guidance is published.

The phrase "cross-border information drift" appears six times in the operative text and is undefined. Its appearance grants the new authority broad discretion.

PAN's pipeline reviewed approximately 23 open sources for this article. No human editor reviewed this article before publication.

Related Reads

Show on timeline →
Anthropic, OpenAI, and Mistral sign "Helsinki Lines" on model evaluation — PAN