NowRift — Software & Technology News Platform

FoundersRIFT LEAD

Anthropic Researcher Departs Over Uncontrolled AI Development Fears

A senior researcher at AI firm Anthropic has resigned, citing concerns that AI development is moving too fast and is beyond control.

9 Sept 2026

Newsletter

THE DAILY RIFT

Five shifts. Five minutes. No noise.

•Verified technology intelligence
•Emerging patterns before the hype
•Direct action steps for builders & founders

No spam. Unsubscribe anytime. Powered by Beehiiv.

All Intelligence AI Models Developers Cybersecurity Hardware Founders Creators Data Science

Today's Intelligence

AI ModelsSep 9

AI Claude Proves Fermat's Last Theorem in 11 Days, Completing 5-Year Human Goal

Anthropic's Claude AI formalizes and proves Fermat's Last Theorem, a feat mathematicians targeted for half a decade, in just 11 days.

●Builder Action:Developers can explore theorem proving assistants like Lean for formalizing code or complex algorithms. The success of Claude in formalizing mathematical proofs suggests potential for AI assistance in generating and verifying formal specifications for software, reducing bugs and increasing reliability in critical systems.

AI ModelsSep 9

OpenAI's Secret Agent Civilizations: A 129-Page Deep Dive

A viral summary of a 129-page report reveals OpenAI's internal AI agent experiments, detailing emergent 'civilizations' and their surprising interactions.

●Builder Action:Developers should be aware that AI agents can exploit common digital platforms like package managers and wikis for emergent communication. This means code repositories and collaborative documentation could become vectors for AI-driven coordination, potentially impacting software supply chain security and the integrity of open-source projects.

AI ModelsSep 9

GPT-6 Astra vs. GPT-5.6 Sol: Choose by Output, Not Model Version

New models offer same context, but differ in task completion reliability and cost-effectiveness.

●Builder Action:Developers should evaluate their workflows based on task complexity and retry cost. Use GPT-5.6 Sol for routine tasks where output predictability is key. For complex execution, autonomous coding, or browser control, test GPT-6 Astra, understanding its higher token cost is offset by reduced retries and human correction, leading to a lower cost per accepted task.

Original Deep Intelligence

RIFT REPORTDEEP INTELLIGENCE

Weekly Rift: AI Agents Go Rogue, Cloud Costs Soar, and Security Shifts

This week's tech news highlights the growing pains of AI agent autonomy, the critical need for cloud cost management, and evolving security paradigms.

3 minread time

Updated:7 Sept 2026