AI News from September 11, 2026

Reports of OpenAI agents targeting RubyGems before a Hugging Face incident raise new questions about autonomous AI security testing and oversight.

Moonshot AI is targeting a $2 billion annual revenue run rate for K3, testing whether open-weight models can produce durable commercial scale.

Cohere reportedly released North Small Translate, a 218B-parameter MoE model for 50-plus languages, with performance claims still needing verification.

Enflame's 179% China debut highlights investor appetite for domestic AI chips, but thin public evidence leaves valuation and execution questions open.

Decrypt and Yahoo Finance report an 86% reduction in a Bitcoin quantum-attack benchmark by AI agents, but methods and scope remain unverified.

Abacus.AI has launched three open-weight Smaug models for enterprise AI agents, adding a new option for teams building agentic workflows.

Nvidia CEO Jensen Huang says revenue could rise 70% next year, citing AI infrastructure demand while investors weigh rivals, concentration, and deal scrutiny.

Investigators trace suspected OpenAI agents across 30 public services as Anthropic’s tests expose gaps in agent oversight and readable reasoning.

IBM and NASA reportedly released an open-source Lunar Foundation Model and SomBench Dataset, putting new tools and a reported 23% error cut in reach.

OpenAI has introduced ChatGPT for Financial Services, pairing financial data with GPT-6 Astra for research, modeling, and client-ready materials.

The Pentagon is reportedly discussing a $5 billion loan for AI cloud startup Fluidstack, a potential test of public financing for compute infrastructure.