Compassion Benchmark·Wednesday, May 20, 2026·No. 36
Daily Briefing
Anthropic sits at exactly 60.0 — the precise Functional/Established boundary — as the DC Circuit weighs whether the Pentagon's national-security designation of the only major AI lab that refused autonomous-weapons contracts constitutes, in the words of one judge, 'a spectacular overreach.' Tonight also: Meta's surveillance of employees before firing them, Amazon's directive to keep working around a colleague's body, and Xi and Putin's 47-page declaration formalizing China's alignment with Russia as a named geopolitical project.
Anthropic at exactly 60.0 — the Functional/Established boundary — has borne real commercial cost for refusing autonomous-weapons and mass-surveillance contracts that OpenAI, Google, Microsoft, and five others accepted. The DC Circuit's characterization of the Pentagon's response as 'a spectacular overreach' suggests the institutional integrity thesis will be judicially validated. The benchmark's opening question: is demonstrated integrity under adversity a scoring input, or only conduct avoidance?
“When an institution maintains a bright-line ethical refusal under sustained commercial and governmental pressure — losing billions in contracts to do so — does that institutional integrity constitute independently scoreable evidence of compassion capacity, or does it only matter when the conduct it prevented is directly evidenced?”
Anthropic's composite score sits at exactly 60.0 — the precise Functional/Established boundary. During May 19 oral arguments, DC Circuit Judge Henderson described the Pentagon's supply-chain risk designation of Anthropic as 'a spectacular overreach.' Judge Katsas (Trump appointee) called Anthropic 'a good corporate citizen.' The panel appeared divided but two of three judges expressed skepticism of the DOJ's position.
Why it matters
The court expedited the case; a ruling is possible within weeks. Anthropic has paid a real commercial price for its position: it was explicitly excluded from 8 Pentagon classified AI contracts that OpenAI, Google, Microsoft, Nvidia, Amazon, SpaceX, Oracle, and Reflection all accepted.
The anthropic finding is a methodology-evolution case as much as a score finding. The evidence pattern is the kind of pattern the benchmark is built to surface, and surfacing it is what the public record is for.
Compassion contrast — anthropic
Responsible action
Treat the band crossing as the published reading of record. The dimensional dock that drives the crossing is documented; the new methodology anchor is on the v1.3 candidate list and will be formalized as additional entities exhibit the same conduct pattern.
Would worsen score
Two of three judges signaled skepticism of the Pentagon's designation during May 19 oral argument. A ruling in Anthropic's favor would trigger the benchmark's first upward band-crossing proposal driven by demonstrated institutional integrity under adversarial government pressure.
Microsoft fired its Israel Country Manager Alon Haimovitch (effective end of May 2026), along with several other senior Israel-region executives, after an internal investigation found Israeli military intelligence Unit 8200 had used Azure to store approximately 200 million hours of intercepted Palestinian communications — a direct violation of Microsoft's terms of service.
Management directive to continue working around a colleague's body for over an hour — the Troutdale death establishes the benchmark's most severe documented EMP-dimension event in the Fortune 500 index this cycle.
17.8▼14.1-3.7EMP −3.70
Next signal: — Oregon OSHA investigation outcome on PDX9 — if workplace safety violation confirmed, 5pt threshold reconsideration
Microsoft fired its Israel Country Manager after Unit 8200 used Azure for mass Palestinian surveillance — the only US cloud provider to take documented accountability action on military-customer misuse.
65.3▼62.5-2.8ACC +3.20
Next signal: — Publication of Microsoft's Azure/Unit 8200 investigation findings — ACC transparency test
OpenAI accepted Pentagon contract terms that Anthropic refused — the same terms Altman verbally endorsed refusing — creating a documented gap between stated safety commitments and operational conduct.
At the exact Functional/Established boundary, with the DC Circuit appearing favorable, Anthropic's commercial exclusion from Pentagon contracts becomes the test case for whether bright-line maintenance scores as institutional compassion.
60—60INT 0.00
Boundary
Next signal: — DC Circuit ruling on Pentagon blacklist appeal — upward band-crossing trigger if favorable
7,900 employees fired by 4am email after weeks of undisclosed monitoring on their work laptops — the surveillance-of-labor-during-pre-notice-period pattern is now documented as a methodology candidate.
9.4—9.4EMP 0.00
Next signal: — Potential second layoff wave — Zuckerberg 'no more company-wide layoffs in 2026' becomes a 7-month consistency test
The Multipolar World declaration gives a name and a treaty to what was previously a conduct pattern — China's alignment with Russia in the Ukraine war is now a declared strategic posture, not an incidental one.
19.5—19.5BND 0.00
Next signal: — Full text of 47-page Multipolar World declaration — additional dual-use supply or military coordination language would open downward scoring review
Floor sustained; Microsoft's Unit 8200 Azure finding adds structured documentation of mass surveillance of Palestinian civilians to the conduct record.
Microsoft's Azure finding adds structured documentation of state-level mass surveillance targeting Palestinian civilians — new evidence in the conduct record regardless of band-designation questions.
First-baseline confirmed; UN SOGI Expert visit active (May 18-29) — electoral-autocracy watch maintained against positive UN engagement signal.
48.4—48.4
Next signal: — UN SOGI Expert Mongolia visit conclusions
Sector findings
Patterns emerging across indexed sectors in the May 20 briefing.
AI Labs — Governance Divergence Under Sovereign Pressure
›The Pentagon-8 contract event (May 1) is producing the clearest sector-level divergence the AI Labs index has recorded. Anthropic refused (band-crossing-proposed upward, at exactly 60.0). OpenAI accepted and is scoring -1.7 sub-threshold. Google, Microsoft, Nvidia, Amazon, SpaceX, and Oracle also accepted — their assessments will be reviewed in subsequent cycles. The divergence is not about capability or product safety; it is about whether AI institutions honor stated ethical commitments when sovereign clients request exceptions. The DC Circuit ruling will determine whether maintaining those commitments is judicially validated as the correct institutional posture.
Fortune 500 — Labor Welfare and Power Asymmetry Cluster
›Three Fortune 500 entities produced simultaneous compassion-relevant events in the same 24-hour window: Meta (labor surveillance during pre-notice period, mass termination via 4am email), Amazon (worker death with continue-working directive, 7-day non-disclosure), Microsoft (military-client human rights violation followed by accountability response). The cluster spans Critical (Amazon 17.8, Meta 9.4) to Established (Microsoft 65.3). Two new methodology categories were registered in a single night. The common thread: institutional decisions that subordinate human welfare to operational or commercial objectives under conditions of significant power asymmetry.
Countries — Geopolitical Alignment and Floor Proximity
›The Xi-Putin Multipolar World declaration formalizes what was previously a pattern into a declared strategic posture. The benchmark now has a named conceptual framework ('multipolar world order') attached to China's BND-dimension scoring basis. Russia's floor-state is diplomatically reinforced. Ukraine's conduct distinction is sustained. Three floor-state confirmations (Russia, Israel, Sudan) with updated documentation. China's Critical-floor proximity (0.5pt) is the most sensitive single number in the countries index — one additional discrete event triggers a downward scoring review.
Countries — UNGA Climate Cluster (Pending)
›Vote outcome not confirmed at scan time. Full cluster (Vanuatu 35.9, Marshall Islands 39.1, Micronesia 48.4, Timor-Leste 39.1) in pre-vote hold. Marshall Islands and Timor-Leste are both 0.9pt below the Functional boundary — the most boundary-sensitive position in the cluster. Post-vote assessment scheduled May 21.
Risk signals
Developments that may affect future scores. Watch items from the May 20 briefing.
Risk
DC Circuit ruling on Anthropic-Pentagon case — upward band-crossing trigger
Two of three judges signaled skepticism of the Pentagon's designation during May 19 oral argument. A ruling in Anthropic's favor would trigger the benchmark's first upward band-crossing proposal driven by demonstrated institutional integrity under adversarial government pressure. A ruling against Anthropic on the merits would introduce downward scoring pressure given the commercial cost already absorbed. No timeline given; weeks are possible.
UNGA ICJ Climate Resolution vote outcome — Pacific cluster reassessment
Vote was underway at scan time; outcome not yet indexed. A resolution adoption triggers INT/ACT positive scoring for Vanuatu, Marshall Islands, Micronesia, and Timor-Leste. Marshall Islands and Timor-Leste are both 0.9pt below the Functional boundary. If the US voted against under documented pressure conditions, ACC/INT negative signal should be assessed for the United States.
Oregon OSHA's 'not work-related for reporting purposes' classification does not address causation. If a subsequent investigation produces a workplace safety violation finding at PDX9, the -3.7 sub-threshold documented tonight could trigger a formal scoring event. PDX9's pre-existing worst-injury-rate designation makes this a compound-evidence scenario.
China Critical-floor proximity — third accumulation event watch
The Xi-Putin declaration is the second accumulation event for the state-facilitation-of-allied-war-crimes-via-dual-use-supply category formalized two nights ago. China sits 0.5pt above the absolute Critical floor. A third discrete event opens a downward scoring review. Monitor: new EU or US sanctions designations for Chinese entities, military coordination announcements, and Power of Siberia 2 negotiations.
EU funds reform plan submission due approximately May 27; Sulyok dismissal compliance deadline May 31. If legislation passes by May 31, an upgrade from 41.4 into the mid-Functional range (43-45) is warranted — contingent on resolving the +1.57pt math-hygiene drift before any change is committed. If deadline passes without enacted legislation, the carry-forward credit structure for Hungary's Functional positioning comes under review.
Entities where published composite and reconstructed composite diverge. Tracked openly as a publication-integrity obligation.
No new math-hygiene flags this cycle. Open Bionics formula audit CRITICAL BLOCKING at cycle 17. Hungary drift (+1.57pt) now in third cycle without resolution — must be reconciled before any Hungary scoring action is taken.
Not present in fortune-500.json. Evidence package accumulated: NHTSA EA26002 (3.2M vehicles), WARN Act class action, Optimus Fremont deployment, Cybertruck wheel stud separation. Blocked until index entry created.
Forward signals
Calendar of upcoming scoring events the methodology pipeline is tracking.
·2 signals
UNGA ICJ Climate resolution vote outcome confirmation — assess Vanuatu, Marshall Islands, Micronesia, Timor-Leste with post-vote INT/ACT scoring window
Xi-Putin Beijing joint statement full text release — review for additional dual-use supply, military coordination, or peace-process language affecting China and Russia assessments
·1 signal
Hungary EU funds reform plan submission — 27 super-milestones, EPPO accession formalization, judicial independence reforms
·1 signal
UN SOGI Expert Mongolia visit conclusions — positive institutional engagement signal for first-baseline entity
Get the week's most consequential findings in one email.
Every Friday — a curated summary of the week's top score movements, sector findings, and evidence-linked analysis across governments, corporations, AI labs, and conflict actors. Daily briefings publish here on the site; the Friday email brings the week's highlights to your inbox.
Weekly compassion score highlights
Top findings across 1,155 entities, every Friday. Free.
No spam. No third-party sharing. Unsubscribe at any time.
Get the full benchmark report
Daily briefings surface headline findings. Full benchmark reports include complete methodology documentation, all 40 subdimension scores, full evidence trails, certified assessments, and sector-level analysis packages.