Google I/O launched Gemini Spark as an always-on AI agent, Meta made its flagship LLM free to use, and India drafted new state AI policy: a remarkably busy week.
Google launches Gemini Spark: an AI that acts for you, not just answers you
At I/O 2026, Google unveiled Gemini Spark, a 24/7 personal AI agent that navigates your digital life on your behalf, transforming Gemini from a chatbot into an active partner. Spark rolls out first to Google AI Ultra subscribers (the new $100/month plan) alongside a redesigned Gemini app with a Daily Brief feature and the Gemini Omni video model.
Why it matters: This is the clearest signal yet that the big labs are moving AI from "answer my question" to "do the thing." If Spark works as advertised, it changes how people use their phones, no more typing searches, you just get results delivered to you.
Meta's flagship LLM is completely free, and leads rivals on health AI
Meta's new Superintelligence Labs launched Muse Spark, its first proprietary large language model under Chief AI Officer Alexandr Wang. Unlike Meta's previous open-source Llama models, Muse Spark is free to any user. It trails GPT-5.4 and Gemini 3.1 slightly overall but leads both by more than double on health-related AI tasks, and matches Gemini on computing efficiency.
Why it matters: A free, capable AI model from one of the world's largest platforms lowers the barrier for anyone (students, researchers, small businesses) to access frontier AI. The health AI advantage is a notable and deliberate bet.
Anthropic shares its unreleased model with Apple, Google, and Microsoft to hunt security bugs
Anthropic launched Project Glasswing, giving select organisations (including AWS, Apple, Cisco, Google, JPMorgan Chase, and Microsoft) controlled access to Claude Mythos Preview, its not-yet-released frontier model. The mission: find and fix critical software vulnerabilities across the industry before malicious actors can exploit them.
Why it matters: Using a cutting-edge AI model to proactively defend the tech industry's biggest platforms is a new approach to safety. It also signals that Anthropic sees Claude Mythos as a major capability step, significant enough to deploy for real-world defence before any public launch.
India writes its own AI rules: NPCI + Nvidia for payments, Rajasthan leads on state policy
India's National Payments Corporation (NPCI) is deploying FiMI (a payments-native sovereign AI trained on Indian financial data) in partnership with Nvidia, already piloted in UPI's Help Assistant at national scale. Separately, Rajasthan's Cabinet approved India's first state-level AI-ML Policy 2026, focused on ethical AI for public services.
Why it matters: India isn't just adopting global AI tools, it's building its own, rooted in Indian data and Indian policy priorities. For anyone in fintech, government services, or public policy, these are the systems that will shape the next decade.
US bipartisan bill targets deepfake distribution and protects AI safety whistleblowers
A new bipartisan US bill would make it illegal to knowingly distribute AI-generated deepfakes, add whistleblower protections for people who report AI safety violations, and require the US to engage in international AI standards bodies. It builds on the Take It Down Act signed earlier this month, which already mandates removal of non-consensual AI deepfakes from platforms.
Why it matters: The US is building a legal architecture around AI harms in quick succession. The whistleblower provision is significant: employees inside AI companies will have legal protection to speak up about safety problems they witness.
EU AI Act transparency rules arrive in August, draft guidance is out now
The EU AI Act's transparency obligations take effect in August 2026, and the EU Commission published draft guidelines this week on what those rules mean in practice. Any AI system that generates text, images, audio, or video for consumers must clearly disclose that it is AI-generated, covering chatbots, image tools, voice apps, and more. Businesses have until August to comply.
Why it matters: This affects any company using AI-generated content in Europe, from startups with chatbots to brands using AI image tools. If you build or market AI products in Europe, August is now a hard compliance deadline, not a distant promise.
When AI acts without being asked, who is responsible?
Google's Gemini Spark and the wider push toward always-on AI agents raise a question that goes beyond features: when AI acts on your behalf without a prompt, who is accountable for what it does? Regulators in the UK (CMA), EU, and US are all beginning to address AI agent accountability this week. The consumer agent market is moving faster than the legal framework around it, and that gap is exactly where problems tend to appear first.
AI's power problem + Parallel raises $230M for AI search infrastructure
Three US senators are investigating Big Tech's role in rising electricity costs; AI data centres already consume over 4% of US electricity and could hit 12% by 2028. Meanwhile, startup Parallel raised $230M at a $2B valuation to build web-search infrastructure specifically for AI agents, a sign that the picks-and-shovels layer of the agent economy is now a major venture bet.