AI Weekly Roundup · Week of July 12, 2026
AI Weekly Roundup · Week of July 12, 2026
A look back at the week’s most important AI developments.
Headlines
1. Moonshot AI Releases Kimi K3: 2.8 Trillion Parameter Open-Source Model Rivals Claude and ChatGPT
Chinese AI lab Moonshot AI released Kimi K3, a 2.8 trillion parameter open-source model that benchmarks competitively against leading US closed models including Claude 4 and GPT-5. The model, which was trained on a cluster of 100,000 H200 GPUs, achieves state-of-the-art results on coding, mathematics, and reasoning benchmarks. Released under a permissive open-source license, Kimi K3 has already been downloaded over 1 million times in its first week, signaling a major shift in the global AI landscape as Chinese open-source models close the gap with proprietary Western systems.
Source: AP News
2. Xi Jinping at WAIC Calls for Global AI Cooperation, 29 Countries Sign Pact
Speaking at the World Artificial Intelligence Conference (WAIC) in Shanghai, Chinese President Xi Jinping called for a unified global framework for AI governance, emphasizing collaboration over competition. The summit concluded with 29 countries signing the Shanghai AI Cooperation Pact, a non-binding agreement committing signatories to shared safety standards, transparency in model development, and the establishment of an international AI oversight body. Notably absent from the signatories were the United States, the United Kingdom, and several European nations, who expressed reservations about the pact’s enforcement mechanisms.
Source: AP News
Model Releases
3. Trump Admin Approves Limited Release of GPT-5.6 Sol and Anthropic Mythos; Pentagon Signs AI Deals Excluding Anthropic
The Trump administration authorized the limited release of OpenAI’s GPT-5.6 Sol and Anthropic’s Mythos cybersecurity model, marking the first government-sanctioned deployments of next-generation frontier models. GPT-5.6 Sol, a lighter iteration of the unreleased GPT-5.6, is approved for use in government and enterprise settings with strict usage caps. Anthropic’s Mythos, trained on classified cybersecurity threat data, received approval for deployment in critical infrastructure protection. However, in a parallel development, the Pentagon awarded major AI contracts to OpenAI, Google DeepMind, and Palantir while conspicuously excluding Anthropic — a decision widely attributed to the company’s vocal warnings about the risks of militarizing advanced AI systems.
Source: AP News
Industry News
4. AI Stock Selloff Intensifies: Nvidia, TSMC Lead Global Semiconductor Rout
A broad selloff in AI-related equities accelerated this week, with Nvidia losing 12% and TSMC dropping 9% amid growing concerns about overcapacity and softening demand from hyperscalers. The selloff was triggered by a combination of factors: reports that major cloud providers are reassessing their GPU procurement timelines, a pullback in Chinese AI chip orders following US export restrictions, and analysts warning that AI infrastructure spending has outpaced actual revenue generation. The declines wiped out roughly $400 billion in combined market capitalization from the semiconductor sector, though some investors viewed the correction as a healthy reset after 18 months of relentless gains.
Source: AP News
5. EU Forces Google to Share Search Data and Open Android to Competitor AI Assistants
The European Union delivered a landmark antitrust ruling against Google, mandating that the company share its search index data with rival AI developers and allow competing AI assistants to be set as default on Android devices. The European Commission determined that Google’s integration of its Gemini AI assistant into Android and Search constituted an abuse of market dominance. Under the ruling, Google must provide third-party AI companies with API access to search ranking signals and allow users to choose their preferred AI assistant during device setup. Google has signaled it will appeal, calling the decision “unprecedented and disproportionate.”
Source: AP News
6. New York Times Accuses OpenAI of Hiding Evidence in Copyright Lawsuit
The New York Times filed a motion in federal court accusing OpenAI of deliberately destroying or withholding evidence relevant to the ongoing copyright infringement lawsuit. The Times alleges that OpenAI failed to preserve internal communications and training data logs after litigation was reasonably anticipated, and that the company’s data retention policies were designed to make it “impossible to reconstruct how copyrighted works were used in training.” OpenAI denied the allegations, stating that it has complied with all discovery obligations and that the Times’ motion is a “baseless attempt to distract from the weakness of their case.” A hearing on the motion is scheduled for early August.
Source: AP News
Research
7. Fujitsu, Nvidia and Japanese Robotics Giants Launch ‘Physical AI’ Initiative
Fujitsu, Nvidia, and a consortium of Japanese robotics companies including Fanuc, Yaskawa Electric, and Kawasaki Heavy Industries announced the “Physical AI Initiative,” a $5 billion joint effort to develop embodied AI systems for manufacturing, logistics, and disaster response. The initiative will leverage Nvidia’s Omniverse platform for simulation and Fujitsu’s supercomputing infrastructure to train foundation models that can operate robots in real-world environments. The partners aim to create open benchmarks for physical AI performance and release a series of pre-trained robot control models by early 2027. Japanese Prime Minister Shigeru Ishiba endorsed the project, calling it “central to Japan’s economic security strategy.”
Source: AP News
Weekly Trend
Open Source Surpassing Closed: A New Phase in the AI Race — The release of Kimi K3 and GLM-5.2 demonstrates that Chinese open-source models now rival top US closed models in coding and reasoning. Meanwhile, the tension between AI safety and military deployment intensifies — Anthropic excluded from Pentagon deals for its safety warnings, yet its strongest cybersecurity model Mythos gets limited government approval. Global regulatory divergence widens: EU pushes for openness, US focuses on security controls, China builds its own governance framework. The semiconductor stock volatility serves as a reminder that AI’s promise rests on expensive hardware.
