Only 18% of banks have AI governance fully embedded, SAS found. I think the other 82% are betting regulators move slower than they do. Where's the gap in your bank's AI governance? The daily read on AI and data: r.7wdata.be/s/hnfScAw

Elvis Presley's estate sued Anthropic over 274 songs in its training data. Same week, LSEG paid $3.535B for data no model can reach. Could you list everything that trained your model? The daily read on AI and data: r.7wdata.be/s/cHUafqv

Four AI rivals each gave away a piece of their agent stack this week. OpenAI, Anthropic, Alibaba and Hugging Face. None of them had to. Which one would you trust running inside someone else's sandbox? The daily read on AI and data: r.7wdata.be/s/VP2zKwT

Music publisher Reservoir sued Suno over 1,050 songs and Anthropic over the lyrics of 853. That's two different AI companies named in one suit, for two different jobs: making music and storing lyrics. The daily read on AI and data: r.7wdata.be/s/AKMELXL

Several US states have put a moratorium on new data centres. 72% of Americans oppose one in their own community (ABC/Post/Ipsos). I think the planning queue now decides 2028 capacity, not the chip queue. The daily read on AI and data: r.7wdata.be/s/Yn5XC9B

Hong Kong's monetary authority now wants a named owner for every AI use case a bank runs. That's the first AI governance rule I've seen that points at a person, not a committee. The daily read on AI and data: r.7wdata.be/s/uYGdV5K

An audit of one codebase found four bottlenecks nobody had flagged in months. Finding them took a careful read, not a smarter model. What's the oldest unreviewed corner of your stack right now? The daily read on AI and data: r.7wdata.be/s/fJnrKQg

Railway sent two warnings this week: check your bill for AI usage you didn't make. This reads like a detection failure, not a billing one. The platform can say where to look, not what was there. Reconciled a surprise AI bill before? Whose job was it?

A new AI health guide tells doctors: this tool does not replace you. I think the regulation disagrees the moment it touches a diagnosis. FDA calls that a medical device, full stop. Who signed off on your last clinical AI tool, you or compliance?

Capital One says its AI rollout now touches more than 3,000 enterprise applications. That's not a pilot number. That's an org that has already decided AI governance is infrastructure, not a committee. What does governance at that scale even look like at your company?

Bloomberg named Tencent, Legend Capital and CPE circling Huawei's Ascend chips. The same four names turn up again around DeepSeek's newest model. Which of Tencent, Legend Capital or CPE surprised you most here?

Ohio State published its guide for how Microsoft's Copilot Cowork agents get paid for. The meter is fine. The reset is the problem: credits pause work until next month, no visible place in line. Who decides which AI task stops first at your company?

Georgia Tech tied 35 CVEs directly to AI coding tools this year. Nobody budgeted review time for code a machine writes instead of a person. If your team ships AI code, who reviews it like a stranger wrote it?

Salesforce, Canva, Braze and Instacart kept showing up together today, across three separate stories that had nothing obviously in common. That's not a coincidence, that's a market consolidating in real time. Watching the same four names collide in your feed?

Insurance just gave old tech debt a new name: agentic debt. Agents piling up with no consistent rule for who decides, who checks, who's accountable. Same trade I've watched for twenty years, new label. Seeing agentic debt pile up in your stack yet?

Bank examiners now have a checklist for your AI's shutdown button. Same week, a piece named the hole in it: the system picks the evidence, the human just signs. Ever produced that trail? Tell me.

PyTorch Foundation just added nine new members this week, Platinum down to Associate. No keynote, no launch party. But that's the floor every model headline stands on. Anyone else watching infrastructure instead of model drops?

Model Context Protocol is quietly bridging AI model development, ontology management, semantic modeling & a dozen-plus other domains teams keep in separate meetings. Everyone's still arguing which model is smartest. Comment 'bridge' and I'll send the protocol connection list.

Kroger cut 3,000 jobs leaning into AI. Singapore filed autonomous robots under security, not safety. Ares bought a 1.9 GW UK power stake. None argued the models are good. They paid for headcount, rules and electricity instead. Comment 'budget' and I'll send the three line items.

Microsoft 365 Copilot had a bug where an unopened email could make it leak confidential data to its sender. The failure has a name from 1988: the confused deputy. We didn't invent a new risk. We just gave it a mailbox. Comment 'deputy' and I'll send the inbox-access checklist.