Sep 29 2026

Anthropic and NIST CAISI assess GLM-5.3 as the most cyber-capable open-weight model

OPENSAFETY

Original reporting

No source was cited for this one in the issue that logged it. The link above searches for it instead of guessing at a URL.

GLM-5.3 built working exploits in 50 of 410 ExploitBench attempts against 56 for Claude Mythos Preview, with safeguard bypass rates of 64% to 100%, about four months behind the US frontier on 's aggregate.

Named in this event

AnthropicZ.aiCAISI

Open calls that touch this

Anthropic holds the Sonnet 5 price increase · 70%, resolves November 30th 2026
Anthropic IPOs before OpenAI · 75%, resolves August 6th 2027
Anthropic keeps Model 2 unreleased through February 2027 · 70%, resolves March 1st 2027
Z.ai releases GLM-5.3's open weights by August 31st · 65%, resolves August 31st 2026
A second flagship API adopts time-of-day pricing · 55%, resolves February 17th 2027
Anthropic runs no production traffic on Fractile silicon in 2027 · 70%, resolves December 31st 2027

Issues that mention Anthropic, Z.ai, CAISI

Around the same time

Sep 29 2026POLICYADOPT

European Commission opens consultation on technology and copyright, including AI

Adigital-strategy.ec

Sep 29 2026MONEYADOPT

EliseAI secures $350 million for housing and health care operations software

Bcommercialobserver

Sep 29 2026POLICY

Executive order directs US agencies to replace "AI" with "Super Intelligence"

search for coverage

Sep 29 2026POLICY

Third Circuit affirms Thomson Reuters' win over Ross, rejecting fair use for AI training

search for coverage

Sep 28 2026MODELOPEN

H Company releases Holo4 computer-use models

Ahuggingface

Sep 28 2026MONEYCHIPS

Nvidia authorizes an additional $150 billion in share repurchases

Aglobenewswire