Afternoon Brief · September 12th 2026

Afternoon Brief, September 12th 2026

Anthropic commits to letting outside evaluators work inside the company. Dario Amodei asks governments to help slow capability development across competing labs. DeepSeek's documentation now keeps V4 Pro available beyond its planned retirement date.

2 min read·Editorial by Nour Haddad

NHNour Haddad
Deployment & the Floor

Anthropic commits to embedded evaluators and calls for slower AI development

Dario Amodei proposed a coordinated slowdown in AI capability development on Saturday and committed Anthropic to giving outside evaluators ongoing access comparable to employees. Their remit would include checking safety practices, reporting incidents and examining training processes before a model is finished.

The rest of his plan needs agreement beyond Anthropic: common standards among companies in democratic countries, government help with constraints, and negotiations with China and other authoritarian governments. Those are proposals, with no international agreement announced. The concrete commitment is outside access; the essay does not establish that evaluators have already started work. SourcesA B

Updated

DeepSeek keeps V4 Pro available beyond its planned retirement

DeepSeek's official documentation now says V4 Pro service will continue after September 14th, with billing unchanged, in response to user demand. That replaces the announced plan to route Pro requests to V4.1 Flash after that date.

Developers relying on Pro can keep requesting it instead of accepting a forced model change. The older Flash names still route to V4.1 Flash, so the reprieve applies specifically to Pro. DeepSeek has inserted the notice into its September 10th changelog entry without a separate revision timestamp; the continuation is confirmed, but the precise time of the reversal is not. SourcesA

Sources

6 citations · 4 primary · 2 secondary

Morning Brief, September 12th 2026