The top things worth knowing about in AI today.
Beijing-based Z.ai, formerly Zhipu, published the full weights of its GLM-5.2 model under an MIT licence on June 17, allowing unrestricted commercial use. The 744-billion-parameter model has a 1-million-token context window and ranks second on the Code Arena coding benchmark, behind only Claude Opus 4.8. For teams that want frontier-level coding performance they can run themselves rather than through a paid API, it is now one of the strongest options available.
Read moreOpenAI published a method called Deployment Simulation that replays past user conversations through a candidate model before release, substituting the new model responses to look for bad behaviour early. Across about 1.3 million de-identified conversations, its estimates of how often a model would misbehave came within roughly 1.5 times the rates later seen in production. It gives labs a way to catch behavioural drift and reward hacking before a model reaches users, not after.
Read moreOpenAI launched a Partner Network on June 16 with a $150 million investment, aiming to certify 300,000 consultants by the end of the year. Launch partners include Accenture, McKinsey, BCG, Bain and PwC, sorted into Select, Advanced and Elite tiers. The move is a bet that the main barrier to enterprise AI is now implementation, finding use cases and redesigning workflows, rather than raw model capability.
Read moreNoam Shazeer, a co-author of the original Transformer paper and until now a Gemini co-lead at Google DeepMind, is joining OpenAI as its lead for architecture research. Google had paid $2.7 billion to reacquire his startup Character.AI in 2024, bringing him back to its AI team. His move underlines how much competition still centres on the small group of researchers who shape how frontier models are built.
Read moreAt its New York summit, AWS introduced Continuum, an agent that finds and fixes code vulnerabilities, and Context, which maps company data into a knowledge graph that other agents can query. Continuum starts in a supervised learn mode and earns the right to act on its own only as customers grant permission, category by category. Both target a common blocker for enterprise agents: safe access to accurate internal data, with clear limits on what the agents can do.
Read moreChina is drafting a plan to spend about 2 trillion yuan, or $295 billion, over five years on a national network of data centres, according to Bloomberg. The blueprint leans on domestic suppliers such as Huawei for at least 80% of chips, largely sidelining Nvidia and AMD. The figure still trails the roughly $725 billion that US firms including Meta and Microsoft plan to spend on AI this year alone.
Read moreFrom June 15, every non-corporate Commonwealth entity must keep an internal register of its AI use cases, each with a named accountable owner, the first mandatory obligation under the updated government responsible-AI policy. Foundational AI training for Australian Public Service staff also became a requirement on the same day. Further obligations, including impact assessments before deployment, follow in December.
Read moreThe Frontier coalition, whose buyers now include Stripe, Google, Salesforce and Anthropic, committed a further $915 million to purchase permanent carbon removal, lifting its total commitments to about $1.8 billion. Anthropic is the first AI startup to join the group, in what it describes as its first climate deal. The spending reflects how the energy and emissions cost of AI data centres is starting to show up in the budgets of major buyers.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.