GPT-5.6 just solved a 50-year math problem, Illinois passed mandatory AI safety audits, and India's Supreme Court banned AI-generated fake legal citations: all in one week.
GPT-5.6 Sol solves a 50-year-old maths problem nobody had cracked
OpenAI launched GPT-5.6 as three distinct models: Sol (high-end reasoning and science), Terra (mid-tier cost efficiency), and Luna (fast and cheap for volume work). On launch week, Sol deployed 64 parallel subagents and produced a verified proof of the Cycle Double Cover Conjecture, an open problem in graph mathematics since the 1970s, in under one hour. xAI's Grok 4.5 also went public the same week, running on a 1.5 trillion-parameter foundation at roughly a quarter of rival token costs, and scoring 83.3% on the Terminal-Bench 2.1 coding evaluation.
Why it matters: When an AI independently solves a problem that stumped mathematicians for 50 years, the ceiling for "what AI can do" just moved. This is not a benchmark score, it is a real scientific result, and it was not announced with much fanfare. Pay attention to the quiet ones.
ChatGPT Work: OpenAI turns its chatbot into an autonomous employee
OpenAI launched ChatGPT Work (its long-anticipated "super app") this week. It connects to Slack, Google Drive, SharePoint, Microsoft Teams, and dozens of other services, and can run multi-step tasks autonomously in the background after you assign them from your phone. It is a direct answer to Anthropic's Claude Cowork agent, launched in January. Currently rolling out to Pro, Enterprise, and Edu users, expanding to Plus and Business users shortly.
Why it matters: This is the shift from AI as assistant to AI as colleague. If you spend your day in productivity apps, this kind of autonomous agent is coming for your workflow, whether your employer introduces it or your competitor does first.
Illinois becomes first U.S. state to require independent AI safety audits
Governor Pritzker signed the Artificial Intelligence Safety Measures Act into law on July 6, setting some of the most comprehensive AI requirements in the United States. Companies building large-scale AI systems must now disclose safety practices, report major incidents, and pass independent third-party audits conducted by qualified experts with no financial conflict of interest. The bipartisan law follows similar legislation in California and New York, signed in late 2025, and is being watched as a template across the country.
Why it matters: Third-party audits are a new accountability layer. Companies can no longer just claim their AI is safe, they have to prove it to an independent reviewer. This template is likely to spread to more states and influence global standards.
India's Supreme Court: AI-generated fake legal citations are now a disciplinary offence
India's Supreme Court set aside court orders that were based on AI-generated content and directed the Bar Council of India to create formal disciplinary rules for lawyers who file AI-hallucinated case citations. The court's order arrives amid a global pattern of lawyers submitting AI-invented judgments as real, courts in the US, UK, and Canada have flagged the same issue. Separately, the Delhi High Court ordered the removal of deepfake posts targeting Rajya Sabha MP Raghav Chadha, one of India's most prominent judicial rulings on AI-generated image abuse.
Why it matters: India's judiciary is drawing a hard legal line on AI misuse in professional contexts. This signals that AI literacy (including knowing when not to trust an AI output) is now a professional requirement, not an optional skill.
PM Modi opens India's first major semiconductor plant, targeting 500 crore chips a year
Prime Minister Modi inaugurated the CG Semi Outsourced Semiconductor Assembly and Test facility in Sanand, Gujarat: India's first large-scale domestic chip manufacturing unit. The facility has already achieved 20 crore semiconductor units of annual production capacity and is targeting 500 crore units. Modi called on India's youth to lead the global revolution in AI and robotics using Made-in-India chips, framing semiconductor sovereignty as core to India's AI future.
Why it matters: Almost all of India's AI compute today runs on imported chips. A domestic supply chain matters for price, national security, and long-term AI independence. This is the first large concrete step in that direction, and it was inaugurated by the Prime Minister himself.
193 countries met in Geneva to agree on how the world should govern AI
The UN Global Dialogue on AI Governance convened in Geneva on July 6–7, bringing together 193 countries to discuss international frameworks for managing frontier AI risks, including catastrophic harms, autonomous weapons, and economic disruption. Simultaneously, the EU activated a July 2026 action plan coordinating AI cybersecurity standards across all member states and commissioned an independent evaluation capacity for frontier AI models before they reach the European market, expected to be operational by 2027.
Why it matters: AI governance is no longer just a national conversation. What gets agreed in Geneva and Brussels shapes the rules every company building AI will have to follow. This multilateral process is slow, but it is now officially running.
The "which AI do I use?" problem is now the real skill to build
In a single week, the public gained access to GPT-5.6 Sol, Terra, and Luna; Grok 4.5; ChatGPT Work; Muse Spark 1.1; and updates across Claude, Gemini, and others. On OpenAI's own benchmark tables, Claude models lead four of the headline evaluations while GPT-5.6 Sol leads the coding agent index. The competition is real, but for most people and businesses, the question is no longer "which AI is the best?" It is "which AI is right for this specific task, at this price, today?" Model routing: the skill of matching the right tool to the right job, is quietly becoming one of the most valuable things you can learn.
EdVisorly raises $13M for AI college advising + Willow.ai launches Atlas-1 speech model
Los Angeles-based EdVisorly closed a $13.3M Series A to scale its AI college advising platform, part of a broader wave of early-stage AI-in-education funding that saw 22 disclosed deals in the past year. Separately, Willow.ai launched Atlas-1, a speech-to-text model aimed directly at ElevenLabs, Deepgram, and OpenAI's Whisper on transcription quality, a market that is quietly becoming very crowded with serious contenders.