The top things worth knowing about in AI today.
At its developer conference Apple confirmed the rebuilt Siri runs on Google's Gemini, but the larger change is that iOS 27 lets apps call third-party models such as Claude and Gemini through a single Swift framework. Developers with fewer than two million downloads also get free access to Apple's on-device models. After two years of Apple Intelligence falling short, the company is now building outside AI into the platform rather than treating it as a stopgap.
Read moreKPMG and Microsoft expanded their deal to roll out Microsoft 365 Copilot across KPMG's entire workforce of more than 276,000 people in 138 countries. The firm will also use Microsoft's new Agent 365 to deploy and govern AI agents across audit, tax and advisory work. It is one of the largest single Copilot deployments announced, and a sign that the big professional-services firms now treat agents as standard tooling rather than pilots.
Read moreOpenAI began rolling out a memory system it calls dreaming, which rewrites what ChatGPT remembers about you in the background instead of keeping a fixed list of saved notes. It can revise an old detail as things change, turning 'you are going to Singapore in July' into 'you went to Singapore in July 2026' once the trip passes. OpenAI reports 82.8% factual recall in its own tests, with Plus and Pro users in the US getting it first. For regular users it means less time correcting stale context.
Read moreNvidia released Nemotron 3 Ultra, a 550-billion-parameter open-weights model published with its weights, training data and recipes under a permissive licence. It is the most capable open model from a US company, though independent testing still places it behind China's Kimi K2.6. For teams that want to run or fine-tune a frontier-class model on their own hardware, the gap between open and closed systems keeps narrowing.
Read moreChinese lab DeepSeek is close to raising $7.4 billion at a roughly $45 billion valuation, backed by Tencent and a state-supported investment fund. The company says it will keep its models open and focus on research rather than near-term products. It is one of China's largest startup rounds and a reminder that well-funded open-weights competition is not confined to the US.
Read moreMeta is testing pricing for a consumer AI agent called Hatch, with reports of a premium tier near $200 a month. In trials the agent handles tasks like drafting emails and building simple software, and Meta plans to run it inside Instagram and its other apps. The price would match the top tiers from OpenAI and Anthropic, a sign that do-it-for-you agents are becoming a paid product category rather than a free add-on.
Read moreThe bulk of the EU AI Act's obligations for high-risk systems take effect on 2 August 2026, covering AI used in hiring, finance, education, healthcare and essential services. Affected providers need risk management, documentation, human oversight and conformity checks, with fines reaching 35 million euros or 7% of global turnover. A May agreement eased and delayed some requirements, but the core deadline stands. Australian firms selling AI into the EU are caught by the rules, so the next eight weeks matter for compliance teams here too.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.