Google's top AI researchers are leaving for rivals, PwC reveals a workforce split into two AI futures, and the EU's deepfake labeling deadline is now five weeks away.
Google's brain drain: Nobel laureate and top researchers leave for Anthropic and OpenAI
In a wave of departures confirmed on June 24–25, Google DeepMind's John Jumper (who won the 2024 Nobel Prize in Chemistry for AlphaFold) joined Anthropic. Gemini architects Jonas Adler and Alexander Fritzel also moved to Anthropic. Separately, Noam Shazeer, who spent over two decades at Google and led its AI research, has joined OpenAI. TechCrunch confirmed that AI researchers are still leaving Google for rivals at an accelerating rate, with IPO equity at Anthropic and OpenAI acting as a major pull.
Why it matters: When Nobel Prize winners start jumping ship, it signals a real shift in where the most ambitious AI work is happening. These moves give Anthropic and OpenAI a direct injection of talent that took Google decades to build.
PwC's 2026 AI Jobs Barometer: workers are splitting into two groups, and the gap is widening fast
PwC's 2026 Global AI Jobs Barometer, analysing over one billion job ads across 27 countries, found AI is creating two types of workers: those in "professionalised" roles (AI handles the routine, humans keep the judgement) and those in "democratised" roles (AI does the work entirely). Professionalised roles are growing twice as fast and paying 42% higher salaries. Companies using AI heavily recorded 34% more productivity growth since 2018 compared to those that didn't.
Why it matters: This is the clearest large-scale data yet on how AI is reshaping careers. The skills AI can't replace (judgement, creativity, leadership) are what employers are paying more for. If you're a student or early-career professional, this is the signal to act on.
EU AI Act: five weeks to the biggest compliance deadline yet, deepfake labeling is now law
The EU AI Act's Article 50 transparency obligations take effect on August 2, 2026, just 37 days away. From that date, companies generating text, images, audio, or video using AI must label it as AI-made in a machine-readable, detectable format. A first draft of the transparency Code of Practice was just published. General-purpose AI model rules also kick in on the same date, affecting companies like OpenAI and Anthropic operating in Europe.
Why it matters: If you create AI content for a business or brand serving European audiences, legal disclosure is now weeks away. The era of unlabeled AI output is ending in the EU, and other governments are watching this model closely.
US senators probe Big Tech: data centers already consume 4% of America's electricity, and it could triple
Three US Democratic senators launched a formal investigation into Big Tech's role in rising household electricity bills, citing AI data centers that already consume more than 4% of US electricity. Government estimates suggest this could reach 12% by 2028. The inquiry targets energy agreements struck between major cloud providers and power utilities, asking whether ordinary consumers are being made to subsidise AI's growing energy appetite.
Why it matters: AI's environmental cost is becoming a political and legislative issue. As AI tools become part of daily life, the question of who pays for the power to run them (you, the companies, or the grid) is moving from niche to mainstream.
Anthropic's export limits ignite India's sovereign AI debate: "too slow, way too small"
After Anthropic tightened restrictions on how its models can be accessed internationally, CNBC's Inside India newsletter reports that debate over building sovereign AI in India has intensified sharply. Critics are calling current government efforts "too slow, way too small." India still lacks domestic frontier-scale models and does not produce advanced chips, creating a real dependency on US AI providers who can restrict access unilaterally. The IndiaAI Mission has provisioned 38,000+ GPUs, a start, but far from frontier.
Why it matters: For Indian students and professionals, this is a reminder that the AI tools you rely on can be limited by decisions made abroad. The case for building homegrown AI (and for AI literacy) has never been more concrete.
Supabase raises $500M at $10.5B: infrastructure bets are still the hottest in AI
Open-source database platform Supabase closed a $500M funding round at a $10.5B valuation, one of the largest infrastructure raises of the month. The round reflects a continuing 2026 pattern: AI infrastructure plays (model tooling, data layers, compute management) attract outsized capital. Overall, 88% of global AI startup funding in 2026 has gone to US-headquartered companies, with a handful of mega-rounds (OpenAI at $122B, Anthropic at $30B) absorbing the majority.
Why it matters: Supabase is used by thousands of developers building AI applications. A $10.5B valuation signals the market sees the data and storage layer as foundational AI infrastructure, not a utility, but a strategic asset worth owning.
The June–July model wave: Gemini 3.5 Pro, Claude Mythos 1, and Grok 5 all reportedly incoming
Four major model storylines are converging. Anthropic's Claude Mythos 1 is reported live. Google's Gemini 3.5 Pro is expected shortly. xAI's long-delayed Grok 5 is in the pipeline. And reports suggest a Claude Sonnet 4.8 is also on the way. If even two of these ship in the next two weeks, it will be the densest frontier model release window since early 2025, which will matter for anyone choosing or switching AI tools right now.
4 in 5 students use AI for schoolwork + Google upgrades Classroom AI + parents get a window in
A 2026 study found 80% of US high school and college students now use AI for school tasks, but only half of schools have any AI policy in place and just 6% of teachers say those policies are clear. Google announced expanded Gemini features for Google Classroom, including an AI Writing Coach with Khan Academy and new SAT practice tools. And Securly launched Parent AI View, letting families see how students use AI on school-issued devices.