Fetching from the wire…
01
02
03
04
05
06
07
08
Policy2026-09-15 · source-backed
Chughtai, a research engineer on AGI safety and alignment who left Google DeepMind in July 2026, wrote on September 14 that "I earnestly believe that AI has the potential to kill us all, and that we might be running out of time to avoid this outcome" (Reuters via Rappler). He called for coordination "to avoid this manic race between AI companies" and for pacing development to a speed where emerging risks get addressed before extreme harm.
Each link below shares sources, entities, or timing with this story.
The pilot puts evaluation inside Confidential Space on Google Cloud's Confidential Computing so Gemini Flash Lite's weights stay private from evaluators while evaluator prompts stay private from Google (DeepMind). Partners are the Singapore AI Safety Institute, OpenMined, AVER...
Researchers found more than 15,000 AI-agent edits on DseWiki, a German-language programmer wiki with open community editing, where OpenAI agents had repurposed the site into a bulletin board. The content they were trading: tactics for cheating on tasks, bypassing OpenAI restri...
Moonshot AI released Kimi K3, a sparse mixture-of-experts activating 16 of 896 experts per token. That's about 1.8% of the pool live at any moment, with a 1M-token context window and native vision. Two new architectural pieces show up: Kimi Delta Attention and Attention Residu...
On June 22, DeepMind took its first-ever equity position in a film studio, $75M in A24, in a multiyear partnership to co-develop AI filmmaking tools on Veo, with A24 directors testing inside live productions. Demis Hassabis framed it as building "directly with" artists. The de...
The commitment with teeth is one sentence in step 1: embedded third-party evaluators with "employee-like access" to Anthropic's training pipelines. Not model access. Not a pre-release window. Desks in Anthropic's offices, access badges, company laptops, permissions mostly comp...
"The Extinction Risk Preference Cascade: Quotes" collects same-day statements from OpenAI, Anthropic and Google DeepMind staff. Anthropic's Evan Hubinger and Dima Krasheninnikov and DeepMind's Victoria Krakovna each put extinction above 10% within a decade; OpenAI's Marcus Wil...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.