OpenAI cut prices with two new GPT-6 models, Claude made a scientific discovery in a real lab, and AI deepfakes are now fooling HR managers at scale.
OpenAI's GPT-6 Sol and Luna: flagship AI at half the price
OpenAI launched two new GPT-6 models (Sol and Luna) priced at roughly half the cost of the previous GPT-5.6 tier. Both are described as "cut from the same cloth" as the top-tier GPT-6 Astra. At the same time, ChatGPT Voice was upgraded to run on GPT-6 and gained new plugin connections, including email and calendar, available on web and mobile.
Why it matters: Better AI at lower cost means more businesses and individuals can access powerful models. The Voice upgrades also push AI further into daily routines, not just typing queries, but speaking naturally with a system that can act on your behalf.
Claude autonomously discovers a CRISPR-like enzyme: AI's first real lab result
Anthropic announced that Claude independently identified a new enzyme system hidden in bacteriophage DNA that works similarly to CRISPR, the gene-editing technology that won a Nobel Prize. This is the first published result from Anthropic's new biolab, where AI runs experiments with real scientific oversight rather than just reading papers.
Why it matters: This is AI making a scientific discovery, not just summarising one. If AI can routinely surface findings like this in biology, medicine, and materials science, it changes the pace of research entirely, and opens doors for scientists who lack expensive equipment or large teams.
OpenAI, Anthropic, and Google are building a shared AI safety standards body
The three biggest AI labs confirmed they have been in active talks for several weeks to create a joint AI safety standards body, an industry self-regulatory organisation that would set shared technical benchmarks and risk thresholds. OpenAI's global policy chief said no antitrust waiver is needed for safety coordination. Over 100 other companies have signed on to the broader push.
Why it matters: For the first time, competing AI companies are aligning on safety rules rather than waiting for governments to impose them. Whether self-regulation is enough is a real debate, but the fact that rivals are talking at all is historically unusual.
AI deepfakes are now passing job interviews. One in three applications may be fake.
New data shows that over one-third of 300 analysed job applicant profiles were entirely fabricated: AI-generated resumes paired with deepfake video interviews that passed real-time screening. Employers in the US are now legally required to notify candidates at least 10 days before using automated hiring tools, but the fake-candidate problem is growing faster than the disclosure rules.
Why it matters: If you are job-hunting or hiring, the process has fundamentally changed. Deepfakes are no longer a future risk, they are an active problem in hiring pipelines today. Knowing this changes how you should verify candidates and how you should present yourself authentically.
Ema raises $77M to automate enterprise workflows with AI agents
Ema, a startup building AI agents that automate business workflows, closed a $77 million Series B led by Creaegis, with Accel, S32, and Prosus also investing. Total funding now stands at $140 million. The round reflects continued investor appetite for AI that produces measurable business results rather than just chatting.
Why it matters: Enterprise AI automation is where the money is flowing in 2026. If your organisation hasn't started evaluating AI agents for back-office tasks, your competitors likely have.
India moves closer to letting AI agents make UPI payments on your behalf
India's payments infrastructure is being extended to allow AI agents to make digital payments through UPI without needing your approval each time. The system is designed for small, repetitive payments with predefined spending limits and rules, think automated bill payments or subscription renewals handled by an AI assistant.
Why it matters: India is embedding AI into everyday financial life at a national infrastructure level. This is not a startup app, it is a government-backed payment rail being retooled for the AI-agent era. The implications for how we manage money and automate daily tasks are significant.
Anthropic's Mythos 5.1 is too powerful for public release
Anthropic launched two new Claude models today: Fable 5.1 for general use, and Mythos 5.1: a more powerful version available only through its trusted access programmes, primarily for vetted cybersecurity work. At the same time, OpenAI confirmed its Astra model meets its own "Critical cybersecurity capability" threshold and is routing its most advanced security features through a limited tester group. Two of the world's most powerful AI models are now too capable for open release. Watch how the line between "public AI" and "restricted AI" shifts as capabilities keep rising.
Xiaomi released MiMo-V2.6 on Hugging Face, including a 309-billion-parameter model under a permissive MIT licence, free for anyone to use. Separately, Alphabet's robotics division Intrinsic released Intrinsic Core, a ROS-compatible open-source environment for building robots, at ROSCon 2026 in Toronto. Both moves continue the trend of large companies releasing research-grade tools that smaller teams can build on.