Europe's deepfake labels are now law, Anthropic just hit its first profit, and OpenAI's IPO is weeks away: three stories that rewrite AI's rules and business model at the same time.
Europe's AI deepfake labels are now the law, and companies face fines up to €15M
The EU AI Act's transparency rules came into full effect on August 2, and enforcement is now active. Chatbots must identify themselves as AI. Deepfakes (AI-edited images, videos, or audio) must carry a visible label. AI-generated content also needs a machine-readable watermark so it can be automatically detected. About 190 companies have already signed the EU's voluntary Code of Practice to show compliance. Companies that ignore these rules risk fines of up to €15 million or 3% of global revenue.
Why it matters: This is the first major jurisdiction to enforce deepfake labeling at scale. Every platform with European users (including apps you use) now has to comply. Other countries are watching this playbook closely. India included.
Anthropic just posted its first-ever profit: $559M in Q2, two years ahead of schedule
Anthropic reported Q2 2026 revenue of $10.9 billion (a 130% jump year-on-year), and its first operating profit of $559 million. The company had previously warned it could lose money for years. Getting here two years earlier than its own internal forecast is a significant shift in how the market is likely to view AI lab economics going forward.
Why it matters: The narrative that AI companies burn endless cash without a clear path to profit just took a serious hit. If Anthropic is profitable, it changes the conversation about sustainability across the whole industry, and for the startups and students watching these companies for career signals.
Meta open-sources Muse Glimmer: a powerful AI that runs on your laptop
Meta announced it is releasing the weights for Muse Spark 1.2 and launching a new family of open-source models called Muse Glimmer, specifically designed to run on a standard laptop without an internet connection. CEO Mark Zuckerberg positioned this as a direct challenge to OpenAI and Anthropic, who keep their most powerful models proprietary. Anyone can now download, modify, and build with these models for free.
Why it matters: A capable AI that runs offline on your own device is a different kind of tool, no subscription, no data leaving your laptop, no usage limits. For students and independent professionals, this lowers the barrier to building with AI considerably.
Maharashtra launches a ₹500 crore AI startup fund. Sarvam AI raises $234M for Indian-language models.
Maharashtra announced its AI Policy 2026, which includes a ₹500 crore AI Startup Venture Fund and 12 new AI incubators across the state. Separately, Sarvam AI (which builds AI models for all 22 official Indian languages) raised a $234 million round, making it one of India's largest AI fundraises to date. Broader data: Indian AI startups raised $1.07 billion across 157 deals in the first half of 2026, up 33% year-on-year.
Why it matters: The money is finally moving toward AI built for India, not just AI adapted for India. Language-native models and state-backed incubators together are a meaningful signal that the ecosystem is maturing. If you are building or studying in this space, the doors are open.
Google's agent protocol joins Anthropic's MCP under a shared neutral foundation
Google's Agent-to-Agent (A2A) protocol (which lets AI agents communicate with each other across different platforms) formally joined the Linux Foundation's new Agentic AI Foundation, the same body that governs Anthropic's Model Context Protocol (MCP). The foundation now has over 250 member organisations including AWS, Microsoft, OpenAI, and Google. This means the two most widely adopted AI agent communication standards are now under shared, neutral governance.
Why it matters: When AI tools from different companies need to work together (your company's AI assistant talking to a supplier's AI) they need shared protocols. Having these standards under neutral governance rather than controlled by a single company reduces the risk of lock-in. It is the plumbing that makes AI interoperability possible.
OpenAI is going public. S-1 filing expected within weeks.
OpenAI's public S-1 filing (the document companies submit before listing on a stock exchange) is expected to land with the US SEC imminently, with a September listing target. The company is reporting roughly $2 billion per month in revenue and a projected $14 billion loss for 2026. Despite the loss, investor appetite is high, and a public listing would make OpenAI the largest AI-focused IPO in history.
Why it matters: Going public changes everything for OpenAI: more scrutiny, quarterly results, pressure from shareholders, and public disclosure of financials. It also marks a shift in AI from a research-driven to a publicly-traded industry. Whatever you think of OpenAI, this is a turning point.
OpenAI's IPO: what happens when an AI lab has to answer to shareholders
OpenAI was founded as a non-profit with a safety-first mission. It is now filing to be a publicly traded company reporting $2 billion in monthly revenue and a $14 billion annual loss. The tension between "move fast and build the best AI" and "answer to shareholders every 90 days" has not been resolved, it is just about to become very public. Watch how the S-1 handles the safety commitments, the governance structure, and the terms for Microsoft's multi-billion dollar stake.
AI education funding hits $95M + US deepfake laws now cover 48 states
AI in education startups raised about $95 million across 11 deals in the first half of 2026, led by language learning and AI tutoring tools. Separately, the US now has sexually explicit deepfake laws in 48 of 50 states (up from 46 at the start of the year) with 58 deepfake-related bills enacted so far in 2026, many with bipartisan support.