The top things worth knowing about in AI today.
On 12 June the US Commerce Department directed Anthropic to suspend access to Fable 5 and Mythos 5 — its two most capable models, released only days earlier — for any foreign national, citing national security. Anthropic could not cleanly separate foreign from domestic users, so it disabled both models entirely for every customer. Officials acted after learning of a method to bypass Fable 5's safeguards and warned Mythos could speed up sophisticated cyberattacks. The episode shows how fast export controls can now reach commercial AI, and that access to a frontier model can vanish overnight.
Read moreMeta is reining in employee AI use after internal consumption hit 73.7 trillion tokens in about 30 days, with internal AI costs on track to reach billions of dollars this year. Staff had been inflating usage — known internally as tokenmaxxing — to climb a leaderboard after AI use was made a performance expectation. Meta is replacing the leaderboard with a central system, AI Gateway, to track and budget spending across teams, with formal allocations from 2027. Even the biggest AI spenders are now treating tokens as a cost to manage, not just consume.
Read moreNvidia has released Nemotron 3 Ultra, a 550-billion-parameter open model that scores 47.7 on the Artificial Analysis Intelligence Index — the highest of any US open-weights model, ahead of Google's Gemma 4 and OpenAI's gpt-oss, but still behind China's Kimi K2.6 at 53.9. Nvidia published the training data and recipes alongside the weights under a permissive licence. Companies that want a capable model they can run and adapt in-house now have a stronger American option, even as the open-weights frontier stays in China.
Read moreAt its Build conference, Microsoft rolled out several in-house MAI models, cutting its reliance on OpenAI. MAI-Code-1-Flash now appears in the GitHub Copilot model picker in VS Code, MAI-Image-2.5 is live in PowerPoint and reaching OneDrive, and Microsoft showed MAI Thinking, a reasoning model. Copilot, now the default AI layer across Windows, Microsoft 365 and Teams, will increasingly run on Microsoft's own models rather than only OpenAI's. For users, that means more AI built directly into the tools they already use.
Read moreShenzhen humanoid robot maker EngineAI has filed confidentially for a Hong Kong listing, three years after it was founded and following a $200 million round that valued it at $1.5 billion. It recently opened a 12,000-square-metre factory it says can build one humanoid every 15 minutes, aimed at security patrols, retail and industrial work. EngineAI joins a wave of Chinese robotics firms, including Unitree, heading to public markets as demand for physical AI grows. The signal for buyers is that humanoid robots are moving from demos toward volume production.
Read moreSix months after a White House order discouraged state-level AI rules, and with federal legislation stalled in Congress, states are regulating anyway. New measures target how chatbots interact with children, how employers use AI in hiring, and what developers must do to prevent AI-caused harm. The result is a growing patchwork that any company operating across state lines will need to track. For businesses deploying AI in the US, compliance is becoming a state-by-state question rather than a single federal one.
Read moreOpenAI is building out advertising in ChatGPT, letting advertisers upload product feeds of up to two million items and adding conversion-optimised campaigns through its self-serve Ads Manager. It has also taken ChatGPT ads outside North America and Australia and New Zealand for the first time, going live in the UK, with Japan, South Korea, Brazil and Mexico to follow. With ads now reaching free users, the assistant is becoming an advertising channel as much as a tool. Marketers will need to decide whether and how to appear there.
Read moreDeloitte's latest enterprise survey finds 61% of Australian companies report efficiency gains from AI, but just 12% say it is already changing their business — half the global figure of 25%. Only 28% have moved 40% or more of their AI pilots into production, and fewer Australian firms plan to lift AI spending next year than peers globally, at 65% against 84%. The gap is less about access to tools than about turning pilots into changed ways of working. For Australian leaders, the task now is moving past experiments to production use that shows up in the numbers.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.