Interactive Visualization Companion

Air Force One, Jensen Huang, and Anthropic's 2028 Memo

A visual reading of Anthropic's May 2026 policy memo as timed corporate advocacy inside a live U.S.-China chips news cycle. The page below stays visual-first: a timeline of the summit moment, a heatmap that separates four different meanings of “distillation attacks,” an evidence chart comparing policy rhetoric to technical literature, and a scenario map of where the memo places causal weight.

Primary Document

Anthropic: 2028: Two Scenarios for Global AI Leadership

Genre Signal

Policy memo with institutional authorship, published on a corporate research site, not a peer-reviewed study.

News Window

Trump state visit to Beijing, May 13-15, 2026; Jensen Huang joins delegation; H200 sales clearance reported during summit.

Core Tension

Compute controls as “master lever” vs a broader ecosystem story of chips, software, recipes, capital, and deployment.

Summit-to-Memo Timing

Timeline density and influence arcs across the May 13-15, 2026 Beijing visit, Nvidia diplomacy, export-control loosening, and Anthropic's memo drop.

Visual mode: event pulse

diplomatic / market event

corporate actor motion

policy memo / narrative intervention

Narrative Pressure Dashboard

Mock intensity gauges for how the episode frames the moment: advocacy timing, export-control divergence, and evidence gap.

Memo genre score

Advocacy > Analysis

Technical split

4 Distillation Modes

Policy mismatch

High in summit window

Empirical certainty

Partial / task-bound

“Distillation Attacks” Disentangled

Hover the grid. The same label is stretched across four very different activities with different norms, transfer potential, and enforcement logic.

Rows are activity types. Columns are dimensions the episode keeps separate: legitimacy, technical transfer, detectability, and policy overreach risk.

From Teacher Output to Student Capability

Step through how compact-model gains can emerge, and where the literature suggests defenses bite hardest.

ordinary compression / post-training

adversarial or ToS-violating path

provider-side mitigation

Deterrability by Mechanism

Toggle between how a hardline policy memo would like the story to read and how the cited technical literature makes it look: partial, expensive, and uneven across tasks.

Mock values are illustrative composites drawn from the episode's reading of trace rewriting, watermark robustness, and compact-model cascade distillation.

Capability Transfer Surface

A matrix showing where unauthorized output mimicry looks most plausible versus where full frontier parity remains weakly evidenced.

low transfer likelihood

moderate transfer likelihood

high transfer likelihood

Bottleneck Allocation

How much causal weight the memo appears to place on each lever, contrasted with the episode's broader ecosystem read.

Scenario Split

Two pathways to 2028: locked perimeter versus porous diffusion.

Interpretive Verdict

The page's closing visual compresses the episode's stance into four weighted balances.

References