As Google I/O wraps, Anthropic calls its next AI the most powerful and dangerous it's ever built, Meta reshuffles 15,000 people for AI, and ChatGPT just got a quiet speed upgrade.
ChatGPT gets a speed upgrade: GPT-5.5 Instant is now the default for everyone
OpenAI has switched ChatGPT's default model to GPT-5.5 Instant: a version tuned for faster, snappier everyday conversations. If you have noticed the tool feeling more responsive lately, this is why. The upgrade applies to all free and paid users automatically.
Why it matters: The AI tool used by hundreds of millions of people just got meaningfully better without anyone having to do anything. Faster, more responsive AI is quietly becoming the new baseline, not a premium feature.
Anthropic: our next AI is the most powerful we've ever built, and the most dangerous
Anthropic is developing a model called Claude Mythos, which the company describes internally as its most capable and most dangerous AI ever built. This comes as Anthropic is reportedly closing in on a $900 billion valuation and preparing for a potential IPO. Co-founder Jack Clark also predicted this week that AI will help humans win a Nobel Prize within 12 months.
Why it matters: The company building powerful AI is also the one saying it is dangerous. That tension is not contradictory, it is the entire point of safety-focused AI labs. But it does raise the question of who decides when a model is safe enough to release.
Meta cuts 8,000 jobs and moves 7,000 employees into AI teams
Meta announced layoffs of roughly 8,000 people (about 10% of its workforce), while simultaneously reassigning 7,000 employees to AI-focused roles. This is not just a cost-cutting exercise. It is a deliberate shift in where the company believes value will be created over the next decade. Similar restructurings are underway at other large tech firms.
Why it matters: This is the pattern playing out across big tech right now: fewer people in traditional roles, more resources flowing to AI teams. Building skills on the AI side of this equation is career protection, not just career advancement.
Google's AI agents are now inside Canva, Instacart, and OpenTable
Building on this week's Google I/O announcements, Gemini Spark (Google's 24/7 personal AI agent) will gain MCP support within weeks, enabling it to work autonomously inside third-party apps like Canva, Instacart, and OpenTable. Separately, Gemini 3.5 Flash is now generally available via the API, running four times faster than comparable models at half the cost.
Why it matters: AI agents moving into the apps you already use daily (not just a chat window) is the shift from AI as a tool you visit to AI as something that works alongside you. This is what "agentic AI" actually looks like in practice.
Sarvam AI is raising up to ₹2,900 crore to build India's homegrown AI
Indian AI startup Sarvam AI is close to completing a $300–350 million fundraise as it bids to build a domestic alternative to US and Chinese AI models. Meanwhile, the IndiaAI Mission is offering 100% compute subsidies to Indian startups building foundational AI models: GPU access at Rs 67 per hour. AI now accounts for 38% of all Indian startup funding in Q1 2026.
Why it matters: India is not just adopting AI, it is trying to build it from the ground up. For students and professionals here, this is a rare window where domestic capital, government infrastructure, and global ambition are converging at once.
EU AI transparency rules arrive in August: AI-generated content must be labelled
The EU AI Act's transparency requirements take effect August 2, 2026. From that date, AI-generated content distributed in Europe must carry machine-readable labels, and users must be clearly informed when they are interacting with synthetic content. A political agreement to simplify other parts of the Act was also reached in May, easing compliance burdens for businesses.
Why it matters: Tools you already use for writing, image generation, and chat will start labelling AI output more visibly. This shifts both expectations and accountability for anyone who creates or shares content online, and is a preview of where other regions are heading.
The "most dangerous AI ever" is also the one everyone is racing to build
Anthropic calling Claude Mythos its most dangerous model yet is not unusual, safety labs do this to justify rigorous testing before release. What is unusual is that this announcement comes alongside a $900 billion valuation, IPO preparations, and predictions of Nobel-winning AI within 12 months. Every major AI lab is racing to build more capable systems while simultaneously saying those systems require more caution. Watch how Anthropic decides to release (or restrict) Mythos when it is ready. That decision will set a precedent others will follow.
GitHub breach, AI education apps raise $38M, and India's new $100M deeptech fund
A trojanized VS Code extension exposed 3,800 GitHub repositories for 18 minutes on May 18, a sharp reminder that AI developer tools carry real security risks. In education, Gizmo (13 million users) raised $22M and Oboe secured $16M to build AI-powered personalised learning platforms. And India's Shastra VC launched a $100M fund targeting AI, defence tech, and space startups.