The top things worth knowing about in AI today.
On 12 June the US government ordered Anthropic to block its two most capable models, Fable 5 and Mythos 5, for all foreign nationals on national-security grounds, and access went dark worldwide within hours, cutting off researchers, clinicians and analysts in Australia and beyond. Anthropic now says the models will return in the coming days, and President Trump told reporters on 18 June that talks are going fine. The episode is a reminder that access to the most capable AI rests on foundations a single government can switch off, which matters for any Australian business building on a US frontier model.
Read moreMicrosoft made Copilot Cowork generally available on 16 June. It is an agent that runs long, multi-step tasks across Microsoft 365 apps and a browser, returning finished work rather than drafts, with nine partner plugins including Harvey, monday.com and Miro at launch. More than half of the Fortune 500 used it during the preview, and it is billed through usage-based Copilot Credits, so the cost scales with how much work you hand it.
Read moreMeta has rolled out a feature called AI Mode in Facebook search, the first time it has replaced the platform standard search with an AI tool. Instead of a list of links, it answers queries using public content from Groups and Reels, powered by the Meta Muse Spark model. Morgan Stanley estimates the feature could bring in more than 10 billion US dollars a year if it reaches a billion users and monetises a tenth of searches, a sign Meta wants a share of the search business Google has long dominated.
Read moreKPMG is expanding its Microsoft partnership to put Microsoft 365 Copilot in front of its entire global workforce of more than 276,000 people, and to deploy Agent 365 to manage and secure AI agents for its clients. The firm is folding the tools into its audit, tax and advisory work, two years after its first Copilot rollout. It is one of the larger single deployments to date, and a signal that big professional-services firms now treat agent governance, not access, as the hard part.
Read moreThe Reuters Institute 2026 Digital News Report, drawn from nearly 100,000 responses across 45 markets, finds social media and video platforms are now the leading way people get news worldwide. AI chatbots remain a small slice: 10 percent use them for news each week, up from 7 percent a year ago and rising to 17 percent among 18 to 24 year olds, but only 1 percent call AI their main news source. Trust is the catch, with just 20 percent trusting news produced by AI chatbots, worth remembering before building any AI summary into a customer-facing product.
Read moreOpenAI released LifeSciBench, a test of 750 real research tasks written by 173 PhD-level scientists across biology and pharmaceuticals. Its best model, GPT-Rosalind, completed just 36.1 percent of them, meaning roughly two in three research-level problems still defeat the most capable systems. For anyone counting on AI to speed up drug discovery, the takeaway is that it is a useful assistant on well-defined work, not yet a replacement for expert scientists.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.