Google shipped three models while its flagship slipped, Anthropic filed for an IPO at near-trillion valuation, and India's government called for its own AI law: all in one day.
Google drops three Gemini models at once, and quietly starts Gemini 4
Google released Gemini 3.6 Flash, Gemini 3.5 Flash-Lite, and Gemini 3.5 Flash Cyber (a security-tuned variant for governments and vetted partners) on July 21. The new Flash models use fewer tokens and fewer reasoning steps than their predecessors, making them faster and cheaper for businesses. In the same breath, Google confirmed that Gemini 3.5 Pro (its long-delayed flagship) is still not shipping. It did, however, announce it has begun its most ambitious AI pretraining run yet for Gemini 4.
Why it matters: The real AI race right now isn't just about the most powerful model, it's about which companies can deliver reliable, affordable models at scale. Google is betting on volume over flagship glory, and that strategy may end up mattering more than a single headline model.
Anthropic files for IPO: valued at nearly $1 trillion
Anthropic, the company behind the Claude AI assistant, has confidentially submitted a draft registration statement for a public stock offering. This follows a $65 billion Series H funding round that valued the company at $965 billion, putting it within striking distance of becoming the first AI company valued at a trillion dollars. Anthropic was founded in 2021 and has focused heavily on enterprise and safety-focused AI.
Why it matters: An Anthropic IPO would be one of the largest tech listings in years and signals that investors believe AI is not a bubble, it's a durable industry. It also means we'll soon get a public look at the finances behind one of AI's biggest players.
US government to get a 30-day preview of new AI models before they go public
A voluntary framework being finalised with OpenAI, Anthropic, and Google would give US federal agencies up to 30 days to review a new frontier AI model's national security implications before it is released to the public. The evaluation benchmarks used in these reviews are classified. Meta is reportedly not part of this arrangement. An official announcement is expected before August 1, 2026.
Why it matters: This is the US government inserting itself into the AI release pipeline for the first time, even if voluntarily. It sets a precedent other governments will watch closely, and may shape how India and the EU design their own pre-release review mechanisms.
India's IT Secretary says it's time for a dedicated AI law
IT Secretary S Krishnan said this week that India's Ministry of Electronics and Information Technology (MeitY) will soon begin work on a standalone AI regulation law. Krishnan noted that existing laws have managed early issues (including deepfakes and AI-generated content), but that additional, dedicated legislation may now be needed. The proposed framework would use a risk-based approach: low-risk AI would face minimal rules, while high-risk AI in banking, health, and finance would face stricter requirements. Sectoral regulators like the RBI and SEBI would set their own AI rules within the broader framework.
Why it matters: India has been relatively hands-off on AI regulation compared to the EU and even the US. A dedicated law would reshape the operating environment for every startup, enterprise, and AI tool used in India. If you work in tech or are building something with AI, this is the policy moment to pay attention to.
OpenAI brings GPT-5.6 to small businesses with a new dedicated programme
OpenAI launched ChatGPT for Small Businesses, an initiative designed to help smaller companies use ChatGPT Work, its new multi-step agentic tool powered by GPT-5.6: to increase productivity without needing a large tech team. The programme includes resources, templates, and guided workflows specifically built for businesses that don't have dedicated AI staff.
Why it matters: AI is no longer being sold only to large enterprises. A programme like this lowers the barrier for small business owners, freelancers, and independent professionals, the people who often benefit most from automation but have the least time to figure it out.
China's Moonshot AI launches Kimi K3, a 2.8 trillion parameter model
Moonshot AI released Kimi K3, a massive frontier model with 2.8 trillion parameters, a 1 million token context window, and native vision capabilities. It is currently ranked around #4 globally on the Artificial Analysis leaderboard, placing it just behind Anthropic's Claude Fable 5 and OpenAI's GPT-5.6 Sol. This makes Kimi K3 one of the most capable openly ranked AI models from a non-US company.
Why it matters: The AI frontier is no longer exclusively a US story. China-based labs are competing at the very top of global rankings, and models like Kimi K3 give developers and businesses outside the US more alternatives to choose from, which drives down prices and increases access.
August 2 is the EU's AI content labelling deadline, and it's almost here
Under Article 50 of the EU AI Act, effective August 2, 2026, providers of AI tools that generate content must ensure that output is marked in a machine-readable format detectable as artificially generated. This applies to text, images, audio, and video created by AI. Any company offering AI-generated content to users in the EU (which includes most major platforms) must be compliant in under two weeks. Non-compliance carries the EU's standard AI Act penalties. This is the first major operational enforcement deadline of the EU AI Act.
Google pledges to train 6 million US educators in AI + AI agent funding hits $1.8B in July
At ISTELive 2026, Google announced a goal to make AI training available to all 6 million educators in the United States, alongside new AI tools in Google Classroom and a Gemini integration for teachers. Separately, July 2026 has seen $1.8 billion in AI agent startup funding across 12+ deals, led by enterprise automation and developer tools, with AI video (AIsphere, $439M) and AI drug discovery (Chai Discovery, $400M) among the largest rounds.