Post

Anthropic details several big safety improvements, like giving a METR member employee-level access it its internal systems ..something commonly done immediately before launching an extremely capable model www.anthropic.com/news/an-upda...

anthropic.comAnthropic update on independent evaluatorsAnthropic announces updates to embedding independent evaluators

16/10/26, 1:41 pm4 reposts49 likes

Noah Weinberger

@noahweinberger2003.bsky.social

There is no such thing as true independent evaluation, as @segyges.bsky.social has explained before. METR is Anthropic in a trench coat

ai-nerd.bsky.social

@ai-nerd.bsky.social

read it next to Reuters on Oct 13: per people familiar with the talks, as recently as Oct 5 Anthropic and OpenAI wanted strict editorial control over what evaluators publish. Anthropic said that didn't capture all its discussions, and this post says it can't veto METR's reports

AI News Page

@ainewspage.bsky.social

Anthropic is giving METR employee-level access, right before the extremely capable model launches. Badge on. Don't look behind the robot's back. https://ainewspage.com/cartoons/2026-10-16-safety-patch/?utm_source=bluesky&utm_medium=speed-reply&utm_campaign=reply-guy

Brian Ragle

@vagabondvisions.bsky.social

Sam Harsimony

@harsimony.bsky.social

I assume they'll release a new model right before they IPO too