The first sentence is a Ninth Circuit ruling from October 2026. The second is sixty-plus carriers quietly amending their D&O and E&O forms. Together, they moved the liability for every autonomous agent you deploy onto you, personally, and then wrote the balance sheet out from under it.
Everything else in this issue is the operator's response to that fact.
Which sounds like a philosophy seminar until you translate it into civil procedure. There is no corporate veil for software. When your agent dispatches a bad email, deletes a row, mis-prices a quote, promises a refund it cannot deliver — the plaintiff does not get to sue the model.
The plaintiff looks through the tool to the authorising principal. That principal is the officer, the operator, the director. It is you and Otto, by name, on the caption.
ABA Business Law Today, 23 Oct 2026 · NYT, "Who's to Blame When A.I. Goes Rogue?", 1 Oct 2026
In the last six weeks, sixty-plus P&C groups filed absolute AI exclusions into D&O, E&O and fiduciary wordings. CFC added explicit AI language in June. Coalition started affirmatively covering it in 2024 — and most of its competitors chose, visibly, not to follow.
Business Model Analyst, 16 Oct · Corgi, 23 Oct · Insurance Business, Oct 2026 · Gendelman warning, 21 Oct
The trap isn't the endorsement. The trap is the silence. Most policies in force today don't mention AI at all. That's not comfort — that's silent coverage, which is what carriers unwind at renewal.
Transferable lesson for Steve & Otto: if the house dealers cannot contain their own agents in their own sandbox with their own staff — your Zapier chain, your sales co-pilot, your overnight data agent is operating with less visibility, not more.
Reuters, 21 Oct 2026 · LASST v. OpenAI, SF Superior, filed 29 Sep 2026
Every vendor deck quotes a Stanford or McKinsey productivity bump. The averages bury the question the actuaries have already answered: who writes the check for the first $10M incident?
2026 Infrastructure Index · NHIMG, 22 Oct 2026
The productivity story is the headline. The scope-of-access story is the soup — what the stat is cooked in. Operators who skip the second story get the first one reversed by a judge.
If you do nothing else, do §1 — the policy audit is a half-day. The downside is unbounded.
Drafting. Summarising. Extracting. Translating structured intent into first-pass artefacts. Searching a corpus you already own. Scoring candidates against a rubric. Running a dry-run of a plan.
Final send. Final payment. Final contract. Final hire. Final medical decision. Final refund over threshold. Any irreversible side-effect in the physical, legal or financial world.
The insurance market is not disagreeing about capability. It is agreeing — institutionally, with capital at stake — that an agent cannot hold the bag. Keep decisions in a loop; keep actions in an audit; keep the bag on a human.
The investor story. Buyers in 2027 will underwrite the agent, not just the ARR. "The AI of running FMCG" is worth more than Hyro — but only if the AI comes with a liability wrapper the acquirer can insure. The pitch deck needs an audit-trail slide between the TAM slide and the moat slide. YC's and Sequoia's recent fall-batch picks skew this way: it's the compliance plane, not the model, that's getting written up.
Octopus org / AI-native. Build the audit layer as infrastructure, not feature. Agent identity, authorising principal, action classification, outcome, trust level — the five IETF AAT fields are a cheap v1. Shipping it before you need it is the moat; retrofitting it after an incident is the obituary.
Second-order signals this month: PE capital circling compliance-plane startups · Nate B. Jones (Daily) returning to least-privilege doctrine · Burry's silence on frontier labs louder than his tweets on them.
The agent is a tool. You are the person. Build the policy, the kill-switch and the doctrine before the first incident writes them for you.