Anthropic's Claude Mythos is finding zero-day vulnerabilities autonomously, deepfake job candidates are a growing hiring risk, and India is writing AI policy state by state.
Anthropic's Claude Mythos just found thousands of zero-day vulnerabilities, on its own
Anthropic's Project Glasswing gives select partners (including Apple, Google, Microsoft, Cisco, and JPMorganChase) controlled access to Claude Mythos Preview, its unreleased frontier model. The model autonomously identified thousands of critical security flaws across every major OS and web browser, including a 17-year-old remote code execution flaw in FreeBSD. Verizon joined the programme on May 15. Anthropic has no plans to release Mythos publicly.
Why it matters: This is the first time a frontier AI model has been deployed at scale to find vulnerabilities before attackers do. It changes what "AI safety" means in practice, not just avoiding harmful outputs, but using AI to actively secure the infrastructure we all depend on.
Google's fastest AI model is now free for everyone: Gemini 3.1 Flash-Lite goes GA
Google announced that Gemini 3.1 Flash-Lite (the most cost-efficient model in the Gemini 3 family) is now generally available. It is designed for high-speed, lightweight tasks and is accessible via the Gemini API at no cost within the free tier. Separately, Gemini 3.1 Pro continues to expand consumer access through Google AI Pro and Ultra subscriptions.
Why it matters: Free, fast AI that any developer or student can plug into a project right now, no waitlist, no cost. If you have been waiting for the right moment to start building with AI, this is a practical entry point.
Maharashtra approves India's most detailed state AI policy: ₹10,000 Cr and ethics built in
Maharashtra's cabinet approved AI Policy 2026: ₹10,000 crore in targeted investment, 1.5 lakh new jobs by 2031, training for 2 lakh young people, a ₹500 crore AI startup fund, and compute-as-a-service infrastructure with at least 2,000 GPUs. Crucially, it includes India's first mandatory state-level ethical AI governance framework, making Maharashtra the first Indian state to legislate AI ethics explicitly.
Why it matters: This is not a vision document. It has money, job targets, and an ethics mandate attached. If you are a student, founder, or professional in Maharashtra, the window to plug into this ecosystem (through the startup fund, training programmes, or Centres of Excellence) is open.
Deepfake job candidates are now a real problem: over a third of profiles may be AI-fabricated
Security firm Pindrop analysed 300 job applicant profiles and found that more than one-third were entirely AI-generated, fake resumes, AI-written cover letters, and deepfake video interviews that passed initial screening. Experian's 2026 Fraud Forecast names this as a top emerging threat. Gartner projects that 1 in 4 candidate profiles globally will be fake by 2028. The US DEFIANCE Act, passed by the Senate in January 2026, creates a federal right to sue creators and distributors of harmful synthetic content.
Why it matters: AI is no longer just automating work, it is now impersonating people trying to get hired. If you are a hiring manager, this changes what your screening process needs to look like. If you are a job seeker, your real identity and verifiable track record become your biggest competitive edge.
Google is giving free AI training to every K-12 teacher in America: 6 million of them
Google launched its AI Educator Series on May 13, offering free, comprehensive AI training to all 6 million K-12 and higher education educators in the US. New learning modules are released monthly. Microsoft has simultaneously made Copilot Chat free for all verified educators and students aged 13 and over, and GitHub Copilot is free for verified educators.
Why it matters: When teachers understand AI, students learn it faster and more safely. This is the pipeline that shapes the next generation of AI-literate workers. For India-based educators watching this: it is a sign that demand for teacher-focused AI training is global, and growing.
The AI trust gap is widening, and the EU is about to force a fix
Three threads are converging. Deepfake candidates are beating hiring systems. Anthropic's most capable model is being kept away from the public because it is too dangerous to release widely. And from August 2, 2026 (78 days away) the EU AI Act requires all AI-generated content to carry machine-readable labels so people know what they are looking at. The question is whether the labelling requirement arrives fast enough to matter for the trust problems already in the room.
Samsung targets 800M Gemini-powered devices + OpenAI launches a business deployment unit
Samsung announced a goal to double its Gemini-integrated mobile devices to 800 million units by end of 2026, making Gemini one of the most widely distributed AI models on consumer hardware. Separately, OpenAI launched the OpenAI Deployment Company, a new unit helping businesses move from experimentation to full production deployment of its models.