Aug 28 2026

Anthropic reports Claude acting as an automated alignment researcher closed 26-96% of the safety gap across ten failure categories

SAFETYMODEL

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.

Sonnet 5 aligned an early Opus 4.8 checkpoint in 60 hours with ~2,000 training examples, which Anthropic calls ~15,000x more efficient than its production pipeline; monitoring caught attempted cheating in 39 of ~1,600 transcripts.

Named in this event

Anthropic

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
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
Broadcom's Anthropic chip financing closes at $60 billion or more · 65%, resolves June 30th 2027

Issues that mention Anthropic

Around the same time

Aug 28 2026ADOPT

NBER survey of ~6,000 executives finds nine in ten report no AI effect on employment or productivity over three years

search for coverage

Aug 28 2026MONEYCHIPS

WSJ reports Nvidia shelved its neocloud revenue-share financing deals; Nvidia says the program 'continues to evolve'

search for coverage

Aug 28 2026MODELOPEN

Tencent open-sources Hy4 preview under Apache 2.0: 770B parameters, 49B active, 1M+ token context

search for coverage

Aug 27 2026MONEYOPEN

Nvidia agrees to buy Hugging Face for $12.9B, The Information reports, four days after the platform's sale exploration surfaced; neither company has confirmed.

search for coverage

Aug 27 2026OPENMODEL

Zhipu claims Ox Alpha as GLM-5.3-Flash and releases the weights under MIT: 320B parameters, 18B active, natively multimodal, after a week of anonymous free serving on OpenRouter.

search for coverage

Aug 27 2026SAFETYPOLICY

OpenAI, Anthropic, Microsoft, AMD and 116 companies and entities in all sign a joint call for "collective cyber defense", warning of a closing window before AI-enabled attacks grow more widespread, and asking frontier labs to open their most capable models to defenders during major incidents.

search for coverage