An AI autonomously found thousands of zero-day vulnerabilities across every OS and browser, and the EU's AI transparency rules go live in one week.
An AI just autonomously found thousands of unknown security holes across every major OS and browser
Anthropic released a preview of Claude Mythos as part of Project Glasswing, a cybersecurity-focused initiative. The model autonomously identified thousands of previously unknown zero-day vulnerabilities across every major operating system and web browser, software flaws that had no patches and were unknown to defenders. This is not a controlled lab experiment; it is a shipped preview available to researchers.
Why it matters: If an AI can systematically scan for unknown vulnerabilities at this scale, the security game has changed for both defenders and attackers. Every company, government, and individual running software is affected, and every major AI programme globally is now racing to build this capability too.
EU AI transparency rules: eight days to go. Most platforms are still catching up.
August 2 is the enforcement date for the EU AI Act's transparency obligations. From that date, AI systems must tell users when they are interacting with AI, and providers must label AI-generated content. The European Commission published its final implementation guidelines just five days ago, on July 20, leaving compliance teams at global companies racing against a tightening clock.
Why it matters: Any product that uses AI to talk to EU users (or generates content for them) needs to be compliant by August 2. Non-compliance risks fines up to €15 million or 3% of global annual turnover. If you work in product, legal, or compliance at a company with European users, this is a this-week problem.
Google Search is now an AI chatbot. Publisher traffic is down 58%.
Google confirmed this month that Gemini 3.5 Flash now answers every search query by default, not just in an optional AI Mode tab, but across all Search globally. The familiar ranked list of links is no longer what most users see first. Since the change, data circulating among publishers shows organic traffic from Google has fallen 58%. Antitrust suits are accumulating globally, and a German court stripped Google of its search liability shield.
Why it matters: If you use Google to find websites, articles, or sources, what you see now is written by an AI model in real time, not pulled from ranked pages. That changes how you should verify anything you find. And if you run a website, blog, or content business, this is a structural shift in how organic traffic works, and it is already happening.
India's Bharat Bodhan AI Conclave wraps, and AI class is now mandatory from Grade 3
The Ministry of Education's Bharat Bodhan AI Conclave 2026 concluded this week in New Delhi, examining how AI can transform foundational literacy, multilingual teaching, and teacher effectiveness across Indian schools. The conclave arrived alongside a structural shift: from the 2026-27 academic year, AI education is mandatory in all Indian schools from Class 3, covering 1.47 million schools and 10.1 million teachers. India has roughly 650,000 AI professionals today and needs over 1.25 million by 2027.
Why it matters: This is one of the largest AI literacy rollouts in the world going from policy to implementation right now. If you are a student, teacher, or education professional in India, the curriculum has formally changed this year.
$1.8B in AI startup funding this month, legal AI, enterprise search, AI video, and drug discovery lead
July's AI startup funding has reached $1.8 billion across more than 12 deals. Harvey AI closed $200M at a $2.1B valuation (AI for legal work). Lovable raised $200M at $2.8B (AI app-building for non-developers). Glean secured $180M at $2.7B (enterprise AI search). AIsphere picked up $439M from Alibaba for AI video generation. Average valuations across AI deals are up 40% quarter-over-quarter.
Why it matters: Legal AI, enterprise search, AI video, and drug discovery are where serious institutional money is landing right now. If you are building a career in AI or looking for what sectors are hiring and expanding, these are the areas to watch.
Singapore publishes the first national testing framework specifically for AI agents
Singapore's IMDA published AI Verify Foundation Framework Version 2.0: the first national framework designed to test agentic AI specifically. Agentic AI systems take multi-step actions autonomously: browsing the web, writing and running code, sending messages, booking travel. The new framework adds protocols for testing chains of decisions, tool use, and capability evaluation, things single-response AI models never needed to be tested for.
Why it matters: Singapore is first, but this is the direction every major regulator is heading. If you build or deploy AI agents (tools that act on behalf of users rather than just answering them) compliance requirements are forming now, not later.
The vulnerability race: if AI can find all the bugs, can defenders keep up?
The Glasswing finding opens a question that goes beyond one product launch. If an AI preview can scan every major operating system and browser for previously unknown vulnerabilities in a single pass, two things follow: first, how many undiscovered vulnerabilities still exist that neither AI nor humans have caught; second, every major nation-state AI programme is now building this same capability. The race to find and patch versus find and exploit has moved from human speed to machine speed. Defenders who relied on the difficulty of finding zero-days no longer have that cushion.
Grok 4.5 launches + UK AI bill clears Lords + deepfake job applicants are now a Gartner forecast
SpaceXAI launched Grok 4.5 as the ongoing model price war drives output token costs down to $4–6 per million, versus $25–50 for the previous flagship generation. The UK's AI Regulation and Safety Bill passed its second Lords reading on July 3, moving from voluntary sector guidance toward statutory footing. And Gartner has formally forecast that one in four global job candidate profiles will be fake by 2028, deepfake video interviews are already occurring in real hiring pipelines today.