America's first state AI consumer law takes effect tomorrow, Anthropic's Mythos 5 is cleared for 100 companies, and Google DeepMind's two biggest names just left for rivals.
US government clears Anthropic's Mythos 5, but Fable 5 is still blocked
The Trump administration approved Anthropic to release its Mythos 5 AI model to roughly 100 vetted companies and federal agencies. Mythos 5 is Anthropic's most capable cybersecurity-focused model. But Fable 5 (its flagship general-purpose model) remains offline after the government pulled it days after its public launch, citing national security concerns about use by foreign nationals.
Why it matters: The US government now effectively decides which AI models businesses can use and when. This is a new kind of power over the tech industry, one that wasn't there six months ago.
GPT-5.6 Sol: OpenAI's newest model is live, but only for approved partners
OpenAI previewed its new model, GPT-5.6 Sol, to a small group of partners individually vetted by the Trump administration. The company says broader availability is weeks away. GPT-4.5 was retired on June 26. Two more variants (Terra and Luna) are also in the pipeline, with Sol positioned as the safety-focused option in the family.
Why it matters: The same government-gating pattern is now applying to OpenAI models, not just Anthropic's. If you use AI tools for work, the approval chain sitting above them is getting longer and more political.
Colorado's AI Act takes effect tomorrow: America's first real state AI consumer law
Colorado's SB 24-205 becomes enforceable on June 30, 2026, requiring companies that use AI in high-stakes decisions (hiring, lending, healthcare) to take reasonable care to prevent algorithmic discrimination. It was delayed once before. The Trump administration has signalled it wants to preempt state AI laws with a federal framework, creating direct legal tension with Colorado's approach.
Why it matters: If you live in Colorado or interact with AI systems for loans, jobs, or healthcare, this law now protects you. For businesses everywhere, it's a signal that "wait for federal guidance" is no longer a reliable strategy.
Google DeepMind's two biggest names just left, for OpenAI and Anthropic
Noam Shazeer, co-lead of Google's Gemini models, is joining OpenAI. John Jumper, the researcher who led the AlphaFold protein-folding project and won a Nobel Prize for it, is moving to Anthropic. Both departures follow a broader wave of talent leaving Google DeepMind, raising questions about whether the lab can stay competitive at the frontier.
Why it matters: The people who built Google's most important AI systems are now building at its rivals. In AI, talent is the product, and this is one of the biggest reshuffles of the year.
OpenAI is preparing to go public, confidential IPO filing in progress
OpenAI is working with Goldman Sachs and Morgan Stanley on a confidential IPO filing, with a potential public listing as early as September 2026. The company is currently valued at $730 billion in private markets. For context: Anthropic's private valuation just hit $965 billion after a $65B Series H, making it the first AI company to surpass OpenAI's private worth.
Why it matters: The AI industry's two leading labs are both approaching trillion-dollar valuations. Once public, OpenAI's financials will be visible for the first time, and the pressure to monetise will intensify.
Deepfake job fraud is now a hiring crisis: 59% of managers say they've seen it
New data from Experian and hiring-tech firms shows 59% of hiring managers now suspect candidates of using AI to misrepresent themselves, and one in three say they've caught a fake identity or AI proxy in an actual interview. Generative AI now enables real-time face and voice cloning during video calls. Experian warns that AI-powered fraud cost US consumers $12.5 billion last year, with job fraud among the fastest-growing categories.
Why it matters: If you're hiring or job-hunting, the rules of trust in remote interviews have fundamentally changed. Verification tools and AI literacy are now table stakes, not optional extras.
Government-gated AI: a temporary measure or the new normal?
Both OpenAI and Anthropic are now releasing their most powerful models only to government-vetted lists of organisations. The Trump executive order allows up to 30 days of national security review before public launch. That framework is voluntary for now, but in practice, labs are complying because the cost of non-compliance is a ban. Watch whether this arrangement gets codified into law, and whether other countries copy the model.
IndiaAI hits 34,000 GPUs + SpaceX renting $1.25B/month of compute to Anthropic
India's IndiaAI Mission has deployed 34,000 GPUs accessible to startups and researchers at subsidised rates, with a target of 100,000 by December 2026. Applications are open for Cohort 2 of the IndiaAI Startups Global acceleration programme in partnership with Station F. Separately, a SpaceX S-1 filing revealed the company is renting 300 megawatts of compute to Anthropic for $1.25 billion per month through 2029, underlining how much raw infrastructure the AI race now demands.