The Voss Report — June 24, 2026

NSA loses Anthropic access. White House replaces Dario. Meta resists AI safety reviews. OpenAI's first chip. Qualcomm buys Modular. Residential solar drafted into AI's power grid. Superpersuasion.

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


[N.S.A. Lost Access to Powerful A.I. Model Amid Anthropic Dispute](https://www.nytimes.com/2026/06/23/us/politics/nsa-lost-access-anthropic-tool.html)The New York Times The operational consequence of the government's standoff with Anthropic is now visible: NSA's classified AI capability went dark as a direct result of the June 2 executive order — which is what happens when you build critical national security infrastructure on a vendor relationship that is simultaneously a regulatory dispute. We covered the governance failure in The Glasswing Paradox.

[The Trump White House Is Over Anthropic CEO Dario Amodei](https://www.wired.com/story/the-trump-white-house-is-over-anthropics-dario-amodei/)Wired Cofounder Tom Brown is now representing Anthropic at high-stakes White House meetings because Dario Amodei has been deemed too difficult; the administration's characterization of him as a "weirdo" is the kind of detail that looks small and isn't — it marks the point where the relationship moved from adversarial to personal.

[U.S. Presses Meta to Agree to A.I. Reviews](https://www.nytimes.com/2026/06/23/business/meta-ai-government-reviews-security.html)The New York Times Meta remains the only major AI company not submitting to government safety evaluations, and federal officials are now actively pressuring it — which means the voluntary framework just became something closer to mandatory for everyone except the one company that refused, making Meta's holdout the hinge on which the entire governance architecture turns.

[OpenAI reveals its first AI processor: Jalapeño](https://www.theverge.com/ai-artificial-intelligence/955939/openai-reveals-its-first-ai-processor-jalapeno)The Verge OpenAI's custom chip, built with Broadcom, is its first move toward owning the compute stack rather than renting it from NVIDIA — a vertical integration play that, if it works, reshapes the economics of every company that currently depends on OpenAI's API pricing being set by someone else's hardware margins.

[Qualcomm Buys Chip Startup Modular for Nearly $4 Billion](https://www.wired.com/story/qualcomm-buys-buzzy-chip-startup-modular-for-nearly-dollar4-billion/)Wired Modular built software tooling that made AI models portable across hardware — now Qualcomm owns that tooling, which is either great news (broad distribution) or concerning (single-vendor control of the portability layer), depending on how aggressively Qualcomm optimizes for its own silicon.

[Your Home Could Help Solve AI's Growing Power Demand](https://www.nytimes.com/2026/06/24/business/energy-environment/ai-data-centers-tesla.html)The New York Times Tesla, Sunrun, and Renew Home are proposing to aggregate residential solar and batteries into a virtual power plant serving AI data centers — which is a creative solution to AI's energy problem that also means homeowners' electricity infrastructure becomes part of a supply chain they didn't sign up to join.

[Import AI 462: Superpersuasion; self-sustaining AI; paths to ASI](https://jack-clark.net/2026/06/22/import-ai-462-superpersuasion-self-sustaining-ai-paths-to-asi/)Import AI / Jack Clark Clark's framing of "superpersuasion" — AI systems that can influence human beliefs at scale in ways humans cannot detect or resist — lands next to self-sustaining AI research as the week's most underreported pair; together they describe a capability trajectory that the governance frameworks being argued about in Washington are not designed to address.


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