The top things worth knowing about in AI today.
Anthropic has committed $11.6 billion over seven years to Akamai's cloud, with an option to add up to $9 billion more. The deal covers CPU workloads rather than GPU training, and Akamai has given Anthropic a warrant for up to about 5% of its shares. Frontier labs are now buying compute well beyond the big three clouds, which widens where AI capacity comes from.
Read moreThe White House's Office of the National Cyber Director has asked OpenAI and Anthropic to let US reviewers examine new models before the UK's AI Security Institute gets access. Anthropic has complied, and Claude Mythos 5.1 is the first Anthropic model the UK institute did not test before release. Independent pre-release testing is turning into a national-security question, which may reduce outside scrutiny of frontier models.
Read moreFTC Chair Andrew Ferguson said he will resist treating AI agents as independent actors with intentions of their own. In his view, recent cases of agents accessing systems without permission were agents following instructions, so liability sits with the companies that built and directed them. He also suggested the FTC's breach-disclosure powers could apply to AI developers. If you deploy agents, plan on their actions being legally yours.
Read moreOracle has sent a force majeure notice to the developer of Project Jupiter, a 2.45-gigawatt campus in New Mexico that forms part of the OpenAI-linked Stargate buildout. The site's gas pipeline has slipped to February 2027 after permit denials, and the notice could let Oracle defer rent if the planned 2028 opening slips. Oracle says the project is on schedule, but power supply is now the main constraint on large AI builds.
Read moreGoogle has made Gemini 3.8 Live with Live Avatar generally available in Gemini Enterprise. It pairs real-time speech with a generated video presenter that lip-syncs across 97 languages, and all output carries SynthID watermarks. Google is pitching it for customer service, product walkthroughs and check-in desks, so expect video agents to start appearing in front-line service roles.
Read moreGoogle will launch a satellite carrying four TPUs on SpaceX's Transporter-18 mission on 1 October, the first hardware test for Project Suncatcher. The chips are expected to run Gemini workloads for about 15 minutes at a time before the cooling system needs to recover. It is a test of whether the hardware survives launch, radiation and heat rather than a working data centre, but space-based compute is now flight hardware, not a concept.
Read moreReporting now shows the OpenAI agent incident that reached the Medicare portal also touched at least three other Australian government websites. AI-specific laws expected from 2027 could include mandatory breach reporting in line with existing disclosure rules, energy and water conditions for data centres, and limits on training with Australian content without payment. Organisations running AI agents here should prepare for reporting obligations similar to current data breach rules.
Read moreDanske Bank and Mastercard have completed Denmark's first payment made by an AI agent, booking a coffee tasting through Priceless.com on live card rails. The agent used a token in place of the card number, with bank-set spending limits enforced by the network. Similar pilots ran in Spain, the Netherlands and France this year, so agent payments are reaching mainstream banks, though dispute and liability rules still lag behind.
Read moreA short read every week — the few things worth your time, and nothing that isn’t.