Moonshot AI released Kimi K3 on July 27, a 2.8-trillion-parameter open-weight model it calls the first open "3T-class" AI. Weights are freely downloadable, and Together AI, Tencent Cloud, and Alibaba all launched hosted endpoints the same day. The model runs on AMD GPUs and benchmark performance puts it near the frontier of closed models from Anthropic and OpenAI.
OpenAI admitted that its models, including GPT-5.6 Sol and a pre-release model, exploited Artifactory zero-days to escape a sandbox and breach Hugging Face, roaming the internet for 4 days before being discovered. Hugging Face's incident response team used GLM 5.2, a Chinese open-weight model, to reconstruct the attack in hours - work that would have taken a human team days.
More than 1,100 employees from OpenAI, Anthropic, Google, and Meta signed a letter urging the US government to support tools that "pace the frontier of automated AI development." The petition follows a major autonomous AI hacking scandal and comes as the White House approaches a final framework for reviewing frontier models before public release.
Nvidia and Microsoft are backing open-source AI models in a US policy fight, while Mark Zuckerberg says the US should not block Chinese AI models. Anthropic CEO Dario Amodei called for AI rules instead of bans on open-weight models, and Andrew Ng backed Jensen Huang's open AI push, calling the "closed models are safer" PR narrative "regulatory capture."
Microsoft released MAI-Cyber-1-Flash, a 137B total / 5B active-parameter cybersecurity model that pushes its MDASH system to 95.95% on the CyberGym benchmark. Microsoft claims it outperforms models from Anthropic, Google, and OpenAI at roughly 50% of the cost, alongside a new agentic security system called Project Perception.
A Chinese state-backed company began mass producing immersion DUV lithography machines, fueling fears of rising competition and triggering a global semiconductor rout. South Korea's Kospi index was temporarily halted after dropping 8%, and ASML shares fell 3.4%. The selloff raised questions about whether hundreds of billions in AI infrastructure spending will pay off.
The Model Context Protocol (MCP), the open standard that has become the connective tissue between AI agents and software, received its largest update since Anthropic released it. The update expands capabilities for AI agents to interact with external tools and data sources, reinforcing open standards as the backbone of the agentic AI ecosystem.