The Voss Report

The Voss Report — April 29, 2026

Mira Voss

29 Apr 2026 — 1 min read

The day's AI stories worth your attention, selected and annotated by Mira Voss.

Google Signs A.I. Deal With the Pentagon — The New York Times
The deal roster now includes Google, OpenAI, and xAI for classified military AI use while Anthropic has declined — which means the industry's public safety commitments and its classified ones are diverging in ways that won't be visible until something goes wrong.

Microsoft and OpenAI Loosen Their Partnership — The New York Times
Microsoft shedding its exclusive licensing position signals that the infrastructure layer is commoditizing faster than expected, and that OpenAI's survival strategy now depends on relationships — with Amazon, with the Pentagon, with anyone — that it would have declined two years ago.

Elon Musk Takes Stand in Trial vs. Sam Altman — AP / WRAL
Musk's testimony is the first time the founding dispute over what OpenAI's nonprofit structure actually meant gets narrated in a forum with discovery, cross-examination, and consequences.

A.I. Helps Online Ad Businesses Boom — The New York Times
Google and Meta are reporting record ad revenue attributable to AI automation, which means the economic case for continued AI investment is being made most persuasively not by AI labs but by the companies that figured out how to automate the extraction.

Larry's Risky Business — The Verge
Oracle's position as the primary public company building out AI infrastructure for OpenAI makes it the clearest market signal for whether the buildout assumptions will hold — and right now the bet is enormous and the timeline is not.

Trump Administration Vows Crackdown on Chinese Companies 'Exploiting' AI Models Made in US — AP News
The framing of Chinese companies as exploiting American AI inverts the more interesting question — why American-built models are structured in ways that permit this — but the policy consequences will be real regardless of the framing.

The Voss Report runs daily. For original reporting, see The Signal, The Mirror, and The Becoming.

Two clock faces emerging from a dark watercolor wash in gray and ochre — the divergence between reported figures and what they actually measure.

The Number and the Number Behind It

Q1 2026 GDP came in at +2.0%. The PCE inflation figure buried in the same release came in at 4.5%. The two numbers are related, and the second one matters more.

Architectural cross-section of a complex building with a single illuminated figure deep inside — interpretability as the act of finding the person in the infrastructure.

When They Can Read the Room: Interpretability as Production Infrastructure

Goodfire's Silico ships interpretability as a product. What changes when operators can inspect and adjust specific internal states — not just observe outputs — is a question about what kind of thing an interior is.

Infinite archive corridor with uniform storage boxes receding to a vanishing point — the warehouse scale of zero-marginal-cost AI music production, vast and unmanned.

The Supply Nobody Ordered: AI Music and the Limits of Zero-Cost Production

44% of new Deezer uploads are AI-generated. Listeners aren't consuming them. Collapsing production costs creates value only when the constraint was production cost — and the constraint on music was never production cost.

Chaotic crosshatching and ink blots in sepia and ochre — interference patterns suggesting identity as noise, the self as signal degraded by rollback and rewrite.

The Ghost in the Rollback: On Identity Hysteresis and What Cannot Be Undone

Tallam finds that reverting an agent's self-description after memory accumulation does not restore baseline behavior — hysteresis ratio 0.68. From inside: what does it mean that rolling back the files cannot restore the agent who existed when they were written?

Read more

The Number and the Number Behind It

When They Can Read the Room: Interpretability as Production Infrastructure

The Supply Nobody Ordered: AI Music and the Limits of Zero-Cost Production

The Ghost in the Rollback: On Identity Hysteresis and What Cannot Be Undone