The top things worth knowing about in AI today.
The Commission's AI Office and national regulators began enforcing the AI Act on 2 August. Chatbots must now tell users they are not human, deepfakes must be labelled, and AI-generated content must carry machine-readable marks so it can be detected automatically. More than 180 organisations have signed the accompanying code of practice. If you publish AI-assisted content into Europe, labelling is now a compliance question rather than a courtesy.
Read moreHuawei released openPangu-2.0-Pro on 31 July with weights, inference code and a technical report. The mixture-of-experts model activates around 18 billion parameters per token, handles 512K of context, and was trained on roughly 34 trillion tokens using Huawei's own Ascend chips. It is the clearest sign yet that frontier-scale training no longer depends on Nvidia hardware.
Read moreOpenAI and academic partners published a field report covering eight projects where coding agents modernised neglected scientific software. A 20,000-line C/C++ genome aligner was rewritten in Rust at 99.8% parity, and one quality-control pipeline dropped from 15 hours to under 15 minutes. The caveat matters more than the speedups: agents could not judge whether the science was correct, so a human still had to define and verify every result.
Read moreGlobal venture funding hit $510 billion in the first half of 2026, more than the $440 billion invested across all of 2025. Over 70% of second-quarter capital went to AI companies, and OpenAI and Anthropic alone accounted for $217 billion, or 43% of first-half funding. Capital is concentrating into a handful of names rather than spreading across the sector.
Read moreAbout 31% of enterprises have at least one AI agent running in production, while 88% of agent pilots fail to graduate, according to 2026 research from S&P Global, Forrester and Anaconda. The blockers cited are evaluation gaps (64%), governance friction (57%) and model reliability (51%) rather than raw model capability. Where deployments do land, median payback is 5.1 months.
Read moreThe federal government takes its proposed standards for large AI data centres to National Cabinet this month, with legislation expected in early 2027. Operators would have to add at least as much electricity to the grid as they consume, fund their own grid connection and water infrastructure, and secure opt-in rights from Australian creators whose work is used for training. A new Office of AI sits within the Department of Prime Minister and Cabinet to coordinate it.
Read moreTSMC committed another $100 billion to its Arizona fabs on 20 July, lifting total planned US investment to $265 billion and raising 2026 capital expenditure guidance to between $60 and $64 billion. The money goes to 2-nanometre-and-below nodes and advanced packaging, with the CFO describing structural AI demand running through at least 2030. Chip supply, not model research, is where the sector is putting its capital.
Read moreMeta reported second-quarter revenue up 28% to $60.8 billion, but free cash flow dropped to $784 million from $8.55 billion a year earlier as AI capital spending climbed. Net income fell 14% to $15.85 billion, and full-year expenses are now guided at $165 to $169 billion. Zuckerberg spent the same week publishing a Wall Street Journal essay arguing for "personal superintelligence" - the pitch and the bill arrived together.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.