Two things moved in opposite directions this week. The price of a capable AI model fell again, and the rules for using these models got more specific, with a November 12 date attached. As AI gets cheaper to use, who carries the obligations that come with using it?
1. Google's Universal Agent Joins the Agent Wars. Google introduced a single agent for work, bundled into subscriptions companies already pay for. The better question than which bot wins: is an agent assigned to every employee what work looks like from here?
2. A $60 Billion Chip Tab Tests the Trillion-Dollar Story. Banks working with Broadcom are raising about $60 billion in chip financing for Anthropic and others. What does that do to the valuation story, and where does the money come from when prices are falling?
3. Haiku 5.5 Removes the Detour Through Open Models. Anthropic's smallest model now costs a tenth of its predecessor's list price. That changes the plan many enterprises had to run internal work on self-hosted open models.
4. Anthropic's New Rules: Skip the Philosophy, Read the Fine Print. The headline was a ban on cruelty toward the model. The part that affects your organization names eleven high-risk areas where a qualified human must review the output.
5. HSBC's Proposed Adviser Cuts Move AI Into Advice. A proposal to cut up to 70% of UK financial adviser roles puts AI into skilled, regulated work. The headcount can shrink; the accountability for the advice cannot.
The Governance Angle. OpenAI starts watermarking text in Europe, fired safety researchers go public, Mistral's largest model tests its limits, and OpenAI's revenue turns out to depend on who is counting.
CIO Corner. From the editor's desk: when a vendor rewrites its usage terms, what do they limit, who answers to the regulator, and who in your organization is watching for the next change?
The 6-Layer Stack. A 3,590-megawatt nuclear agreement, a guilty plea in a chip-export case, and a price list that now depends on how much you send. Six signals from energy to applications.
Agent 101. Retrieval is what your agent looks up. Memory is what it writes down. They need different rules, and only one of them creates new records about your people.
Models got cheaper and the rules for using them got more specific, in the same seven days.
See you next week — still watching, still distilling.
— The Distilled AI Digest Team
