Anthropic eyes an October IPO, EU deepfake labeling becomes mandatory next month, and Gemini 3.5 Pro lands in two days: July 15 was the calm before a very loud week.
Anthropic is preparing for an IPO, as early as October 2026
Reports say Anthropic is drafting its S-1 filing for a public stock listing, targeting October 2026 at the earliest. The company is also in active talks with Samsung to co-develop a custom AI chip, reducing dependence on third-party suppliers.
Why it matters: If Anthropic goes public, it becomes the first frontier AI lab to list on a stock exchange. That changes how the company is governed, who it answers to, and how it funds its research. Worth tracking closely.
GPT-5.6 Sol now runs at 750 tokens per second, for everyone
OpenAI's GPT-5.6 Sol, running on Cerebras hardware, is now available at speeds up to 750 tokens per second, far faster than any previous consumer AI product. To put that in context: a typical AI reply that previously felt instant now loads in a fraction of a second.
Why it matters: Speed unlocks use cases that slow AI cannot do, real-time translation, rapid legal research, live coding assistance. This isn't just faster ChatGPT; it's a different category of tool.
Gemini 3.5 Pro arrives Thursday, and Xi Jinping opens the World AI Conference the same day
Google's next flagship AI model, Gemini 3.5 Pro, is widely expected to launch on July 17 with a 2-million-token context window and a new "Deep Think" reasoning mode. The same day, China's President Xi Jinping is set to attend the World Artificial Intelligence Conference in Shanghai, his first personal appearance at the event since it began in 2018.
Why it matters: Two signals on the same day: the global AI model race escalating, and China signalling AI is now a matter of presidential priority. Both shape what comes next.
EU AI Act: Every AI-generated image, audio, and video must carry a label: August 2 deadline
Starting August 2, 2026, the EU AI Act requires all AI-generated content to carry a machine-readable label. Platforms must also disclose to users at first interaction when they are seeing or hearing synthetic content. This covers images, videos, audio clips, and even AI-generated voices in real time.
Why it matters: This is the most sweeping content-transparency rule any major government has passed. If you use or share AI-generated media, you'll start noticing labels and disclosures on European platforms in two weeks.
India's deepfake takedown law is being enforced: 3 hours to remove, 2 hours for NCII
India's IT Rules 2026 (notified February 10, operational February 20) require platforms to remove flagged AI-generated deepfakes within three hours of a government notice, and within two hours for non-consensual intimate imagery. Courts invoked these rules twice in the first two weeks of July: the Delhi High Court ordered takedown of deepfakes targeting a sitting MP, and the Madras High Court ordered police to act on deepfakes targeting a private individual.
Why it matters: India moved from announcing rules to actually enforcing them. Courts are now using IT Rules 2026 in live cases. Platforms operating in India need to have removal workflows ready, and individuals now have a legal path to act fast.
NVIDIA releases Nemotron 3 Nano Omni: one open model that sees, hears, and reads
NVIDIA's new Nemotron 3 Nano Omni is a 30-billion-parameter open model that handles vision, audio, and language together, no separate models needed. It delivers up to 9 times more throughput than comparable open-source alternatives, and developers can download and run it themselves without a commercial licence from NVIDIA.
Why it matters: The most capable open multimodal models have previously required large budgets and cloud infrastructure. This lowers the floor significantly for researchers, students, and developers building AI tools that go beyond text.
India's AI sovereignty challenge: caught between the US and China
A Bloomberg analysis this week maps out India's uncomfortable position in the AI race. India's 2,117 Global Capability Centres employ 2.36 million professionals and clock nearly $100 billion in revenue, but the country's AI infrastructure depends heavily on US-origin chips and cloud platforms. As US-China AI rivalry intensifies, India is trying to build sovereign compute and policy independence without alienating either side. The IndiaAI Mission's 38,000 GPUs are a start, but the gap between ambition and independence is wide. Watch how India walks this tightrope over the next six months.
BMW bets $300M on agentic AI startups + Anthropic's cybersecurity program triples in size
BMW i Ventures announced a $300M fund targeting early-stage startups in agentic AI, physical AI, and industrial automation, a sign that traditional manufacturing is done watching AI from the sidelines. Separately, Anthropic's Project Glasswing, which deploys a restricted AI model to find and fix security vulnerabilities in critical software, has grown from 50 to 150 partner organisations across 15 countries.