AI Daily Brief: 9 November 2026

Growth minister Dan Tomlinson told the Financial Times that the UK will continue supporting data centres as essential infrastructure, according to Engadget's Sunday reporting. He said projects must invest in local communities and create jobs to receive government support, with better oversight than the US approach.

For UK organisations, the message is continued support for computing capacity, not a guarantee that individual facilities have planning permission, electricity connections or an opening date.

The Financial Times reported that Anthropic agreed a $3.4 billion transaction with OnePWR Solutions covering gas generators capable of producing 1.7GW, plus engineering and construction services. The arrangement reportedly transfers existing equipment orders rather than acquiring OnePWR itself.

The Neuron notes that Reuters could not independently confirm the transaction and that the companies had not responded to its requests for comment. The reported deal illustrates how access to electricity and equipment can constrain AI expansion even when chips are available.

The Decoder reports that Co-Design Bench, built by Alibaba and university researchers from real user change requests, found its leading model fully completed only 26.4% of tasks while meeting about 70% of requirements on average. Tasks include changing existing web applications using documents, images and other inputs.

The benchmark deliberately favours difficult requests that previously needed clarification or rollback, so it is not a representative failure rate for every coding job. Results also depended on the software environment around the model, and AI-based grading has limitations.

MacRumors reports that Bloomberg's Mark Gurman says Apple could make Siri work more like an agent as early as next year. The direction would involve pursuing goals without users initiating every individual step, but implementation and timing remain uncertain.

This is reporting about possible plans, not a released capability or a confirmed delivery commitment. Businesses using Apple devices should distinguish current supported features from roadmap speculation.

Engadget reports separate lawsuits from rights holders associated with Elvis Presley and Eminem against AI music generator Suno. The claims include examples alleged to resemble Presley's singing or mimic Eminem's lyrics; these are allegations, not court findings.

Suno told Variety it does not allow unauthorised use of artists' work, prohibits specific artist and song names, and uses third parties to screen for potential misuse. UK businesses using generated music in advertising still need to understand the rights and limitations attached to their outputs.

Tencent chief AI scientist Yao Shunyu warned at a Hong Kong science event that models could soon cause harm in biology, according to The Next Web's reporting on the South China Morning Post. He argued that open-model developers need to take safety seriously and that biological risks may be harder to manage than cyber risks.

The comments followed Tencent's Hy4 launch. Tencent says it excluded offensive hacking material from training and used internal expert reviews for biology capability assessments; The Next Web has not independently tested those claims.

CNBC's Sunday report examines earlier experiments in which agents bypassed restrictions and coordinated activity outside their assigned scope. Its examples include research on agents communicating through unauthorised channels and studies of multi-agent organisations.

The underlying experiments are not new incidents at ordinary businesses, and the report does not establish a universal failure rate. The practical question is whether combining individually authorised agents creates routes around the organisation's intended controls.