OpenAI's audited 2025 accounts leaked, and the number everyone will repeat is a £28.8 billion ($38.5 billion) net loss. The number that matters is smaller: roughly £31.1 billion ($41.55 billion) of that loss is a one-off, non-cash charge tied to converting from a nonprofit to a for-profit structure, not money spent. Strip it out and the operating loss was £15.6 billion ($20.92 billion) on revenue of £9.8 billion ($13.07 billion), itself up from about £2.8 billion ($3.7 billion) the year before. Revenue more than tripled.
A vast net loss can hide fast revenue growth on far less cash burn than the headline implies, or it can mask genuine overspending, and the leak lets anyone judge for the first time. The AI boom now has audited numbers attached, which turns a question of faith into one that risk and treasury desks can price.
The US Treasury launched Trump Accounts around Independence Day. Every eligible child born between 2025 and 2028 receives a £747 ($1,000) government contribution, and family, friends and employers can add up to £3,740 ($5,000) a year, auto-invested in a low-cost index fund. BNY runs the national infrastructure; Robinhood builds and operates the front end. It is a state-seeded, fintech-operated investing rail at national scale: a customer relationship that begins at birth, held by a broker rather than a bank.
A child with a funded account from year one arrives at adulthood already inside one firm's system, with a default provider and a decade of contributions. The design is a customer-acquisition engine aimed at a generation before it opens a first current account, a near-term variable for any incumbent's deposit and youth-investing franchise.
Google filed a civil suit against a phishing-as-a-service operation it calls the Outsider Enterprise, whose members prompted Gemini to generate the HTML for brand-impersonation pages. Subscriptions started at about £66 ($88) a week. Google cites more than 9,000 fake sites and over a million fraudulent URLs, with 2.5 million scam texts sent to Android users in a single two-week span. A model-maker suing to shut down criminal use of its own tool is a new move.
It puts the provider, not just the victim bank, on the enforcement line. The impersonated brand in most of these scams is a bank, and the cost of AI-made fraud is now measured in billions aimed at bank-branded channels, a direct charge on trust in the sector's own alerts.