AI is scaling faster than organizations can control
THE SO WHAT
The constraint has flipped from ‘how do we get AI adopted’ to ‘how do we keep up with what we’ve already turned on’—shadow agents and unmanaged prompts are now a governance problem, not a curiosity. This week, inventory where models are actually running, who owns them, and what data they touch before you add another AI initiative to the roadmap.
READ THE SOURCE
MORE FROM THE WIRE
Applied AIRunable hits $21M to bet AI agents can go from building businesses to growing them
If 60%–70% of a trillion-plus tokens in 90 days are from paying customers, agentic workflows are already past toy status for a subset of operators. Treat this as a prompt to quantify your own unit economics on agent usage — tokens, supervision time, and downstream revenue — before this pattern hardens into a new cost baseline.
Applied AIThe SEC has subpoenaed four banks over Situational Awareness
Regulators pulling lender communications and trade timing on an AI-heavy fund is a warning shot — leverage plus opaque model-driven strategies is now a live enforcement vector. If you’re running AI-informed trading or structured products, tighten documentation around model use, risk limits, and lender disclosures before subpoenas arrive.
Applied AIGoogle’s Gemini AI Is Changing How Deutsche Bank Assesses Credit Risk
If a Tier 1 bank is putting Gemini into credit risk assessment, model outputs are now upstream of capital allocation and regulatory exposure. Any financial institution experimenting with LLMs should lock in audit trails, challenger models, and clear human override points before these systems touch real balance sheet decisions.
Applied AISources: Moonshot AI is in early talks over revenue-sharing agreements with Microsoft, Amazon, and Google to host Kimi K3, and is seeking up to a 30% share
A Chinese model provider pushing for up to 30% of cloud revenue on Kimi K3 usage shows how foundation models are trying to claw value back from infra. If you’re a heavy model consumer, expect pricing and lock-in to get more complex — multi-model, multi-cloud strategies become less optional as these negotiations harden.