Compassion Benchmark

Viewing archive: Apr 29

Back to latest
Live · Wednesday, April 29, 2026Issue No. 29.04

ACB Daily Briefing

Daily intelligence on AI, institutions, power systems, and measurable human impact.

We translate global events into compassion risk signals, institutional accountability insights, and early warnings for systems that shape human lives. Every finding is grounded in primary-source evidence — litigation records, regulatory filings, investigative reporting, and international legal instruments.

Tonight's pipeline
Scanned
1,155
Assessed
16
Score changes
0
Confirmed
15

How this briefing works

Research agents scan all 1,155 benchmarked entities nightly for new evidence. Flagged entities receive full 40-subdimension assessments. Score changes are proposals, not automatic updates — a human analyst reviews each before published scores change.

Today's top signals

The most material cross-entity intelligence from tonight's research. Each signal is grounded in primary-source evidence and sorted by severity.

Critical severityMethodology·Signal 01

Floor-Limitation Cluster — CRITICAL

What happened

Five entities at or near scoring floor (xAI 2.2, Palantir 6.6, Sudan 0, South Sudan 0, Israel 8.8) with substantial new evidence each night that cannot register through current methodology. Gap now 8 consecutive nights overdue for methodology-note resolution.

Why it matters

Evidence documented but structurally unexpressible. Transparency gap is itself a methodology statement if left unaddressed.

What's next

URGENT: Human decision required on methodology-note framework (permanent-floor designation, cumulative evidence tier, or sub-score harm-scale qualifier). Three options fully specified in PENDING_CHANGES.md.

High severityAI·Signal 02

AI Labs — Safety Incident Cluster

Dimensions implicated
SYSnegative pressureINTnegative pressureACCnegative pressureAWRnegative pressure
What happened

Anthropic Mythos Preview breached within 14 hours of restricted launch via guessable URL and improperly-scoped vendor credentials (April 21, Bloomberg/SiliconAngle/TechCrunch). Anthropic investigating as of April 22.

Why it matters

Separately, Musk v. Altman trial Day 3 (April 29) continues with Musk retaking stand; $150B potential damages; nonprofit-to-PBC governance implications for all frontier AI labs. EU AI Act full enforcement 95 days away (August 2, 2026); only 8 of required member-state contact points designated as of March 2026.

What's next

Monitor Anthropic breach investigation resolution. Mandatory OpenAI re-queue post-verdict (~May 21).

Medium severityAI·Signal 03

Fortune 500 — AI Labor Cascade

Dimensions implicated
EMPwarningINTwarning
What happened

Anti-double-count applied for Microsoft and Meta AI (April 23 announcements absorbed April 28). 92,000+ tech workers laid off YTD 2026.

Why it matters

Cascade confirmed at Amazon, Oracle (index gap), Meta, Microsoft — all citing AI automation as explicit enabler. Content moderation reduction at Meta has downstream user-safety implications not yet scored.

What's next

Pipeline needs EMP voluntariness modifier before next full F500 rotation sweep. Oracle index gap remains unresolved.

Affected entities
Low severitySovereign·Signal 04

Countries — Rotation Baseline Documentation

Dimensions implicated
ACCwatchINTwatchBNDwatchAWRwatch
What happened

Five countries received first-ever agent assessments as T3 rotation: Belgium (77.9), Singapore (62.2), Portugal (65.6), Japan (65.6), Australia (62.5). Singapore April 2026 OHCHR drug-execution moratorium call is the strongest rotation-cycle negative signal.

Why it matters

Australia January 26 UN HRC review with 40-country call to raise criminal-responsibility age reflects sustained pattern. All five confirmed at live baselines (scanner baselines diverged from live for all five — reconciliation recommended).

What's next

Scanner baseline reconciliation pass recommended before next rotation cycle.

Floor designations

·6 entities at composite 0 with documented evidence pattern

Composite scores resolving at zero — methodology disclosure

These entities have all 8 dimensions resolving at the lowest behavioral anchor (1.0/5.0) across multiple assessment cycles. The composite is zero by formula; each entity page surfaces the documented evidence pattern. Floor designation is reversible when functional improvement is evidenced. Read the methodology.

Source intelligence

Primary-source alerts from overnight scanning. Each alert is linked to original regulatory filings, court records, investigative reports, and international legal instruments.

AI Labs — Safety Incident Cluster

Floor-Limitation Cluster — CRITICAL

Fortune 500 — AI Labor Cascade

Countries — Rotation Baseline Documentation

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.

Scores confirmed

Entities where research found published scores remain accurate. Confirmations are documented evidence, not silence.

EntityIndexBandPublishedAssessedDeltaDateFinding
AnthropicAi Labsestablished61.661.60Mythos Preview Discord-group breach (April 21, Bloomberg/TechCrunch) is the most significant new SYS/ACC negative signal: unauthorized access within 14 hours via guessable URL and improperly-scoped vendor credentials. Offset by voluntary too-dangerous-to-release classification (positive AWR/INT) and April 22 DC Circuit filing. Sub-5pt net effect; composite 61.6 Established holds. Material breach documented for record.
OpenAIAi Labsdeveloping27.527.50Trial Day 3 (April 29): Musk retook stand; Jared Birchall next witness. Two surviving claims: unjust enrichment and breach of charitable trust. ~$134B sought; jury advisory only, judge decides. SB3444 committee outcome unconfirmed. Verdict approximately May 21 liability phase. Post-verdict mandatory re-queue. Composite 27.5 Developing holds.
Palantir AIAi Labscritical6.66.60Floor-limited confirmation. Karp 22-point Technological Republic manifesto (32M views, April 20) sparked global technofascism backlash. Wired Slack internal employee-ethics rupture documented. UK Parliament floor characterization as dreadful and shameful. Met Police probe of Palantir-flagged officers. Composite 6.6 cannot register further movement. Methodology-note candidate.
xAI/GrokAi Labscritical2.22.20Floor-limited confirmation. Hard floor maintained. No acute April 25-29 net-new evidence beyond prior cycle absorption. Continuing multi-country regulatory accumulation pattern (ICO UK, NY AG, EU Commission DSA proceedings). Floor-limitation methodology gap overdue 8 nights. Methodology-note candidate.
Meta AIAi Labsdeveloping29.429.40Anti-double-counting applied. April 23 10% layoff (~8,000 employees, effective May 20) already absorbed in April 28 confirmation. No new April 24-29 net evidence. Composite 29.4 Developing holds.
DeepMind/GoogleAi Labsestablished65650No new acute April 24-29 evidence beyond April 17 FSF v3 absorbed in prior cycle. AI labor cascade sector alert applies as contextual only. Boston Dynamics Atlas factory deployment ramp (Hyundai/Google DeepMind, 2028 target) noted. Composite 65.0 Established holds.
SudanCountriescritical000Floor-limited confirmation. War year-4 anniversary (April 15) anchored extensive in-window reporting. April 27: RSF holding 1,470+ civilians, 907 military, 20 doctors in el-Fasher with cholera outbreak. $3B aid plan critically underfunded. 65% of population in urgent humanitarian need. Composite 0 holds at absolute floor. Methodology-note candidate.
IsraelCountriescritical8.88.80Floor-limited confirmation. 37 NGO license non-renewals; aid down 37% Jan-Mar vs. ceasefire period; UN famine confirmation in Gaza governorate; one-third of health facilities threatened with closure. April 28 UNSC ceasefire briefing. Partial mitigation: Zikim Crossing aid surge. Composite 8.8 holds at near-floor. Methodology-note candidate.
MicrosoftFortune 500established66.466.40Anti-double-counting applied. April 23 voluntary buyout (~7%/8,750 employees) absorbed in April 28 confirmation. No new April 24-29 net evidence. Composite 66.4 Established holds.
HoustonUs Citiesdeveloping35.235.20Anti-double-counting applied. April 22 13-4 council vote gutting ICE cooperation ordinance was basis for April 27 rebaseline at 35.2. Civil rights backdoor repeal framing already absorbed. No new April 23-29 net evidence. Composite 35.2 Developing holds.
AmazonFortune 500critical17.817.80April 20 WARN Act investigation (TMB8 Homestead FL) and April 21 NELP Flex brief (systematic wage theft, unsafe work speeds, racial discrimination) documented. April 20 CA AG price-inflation evidence absorbed yesterday. No threshold-crossing event. Composite 17.8 Critical holds.
TexasUs Statescritical14.114.10Floor-pressured confirmation. Abbott $114M Houston funding coercion, HRW Operation Metro Surge context, Amnesty Texas-specific petrochemical and immigration enforcement concerns. Composite 14.1 Critical with EQU/ACC/INT at 1.3 sub-band ceiling. Floor pressure note: composite may not survive next cycle with continuing evidence accumulation.
DeepMind/GoogleAi Labsestablished65650T2 confirmation. No new acute evidence beyond April 17 FSF v3. Composite 65.0 holds.
United StatesCountriesdeveloping25250Amnesty April 2026 World Report documents migrant clampdown including practices amounting to torture and enforced disappearance. HRW Operation Metro Surge unlawful deaths (Renee Good, Alex Pretti) by armed federal agents. Composite 25 Developing already deeply reflects accumulated evidence. DATA INTEGRITY FLAG: user-input baseline 35.5 diverges from live 25.0; live used per protocol.
MinnesotaUs Statesexemplary84.484.40HRW Operation Metro Surge unlawful deaths attributed to federal actors, not state government. Per attribution methodology, state composite not penalized for federal enforcement conduct within state borders. State EQU at 3.8 already reflects federal-context vulnerability. Composite 84.4 Exemplary holds.

Key highlights

Editorial-level findings from the Apr 29 research cycle.

01

Second consecutive zero-proposal night: pipeline operating on clean confirmed state for all assessed entities; rotation-state drift gap substantively closed.

02

Anthropic Mythos breach (April 21): most significant new material evidence registering as confirmation rather than proposal — 14-hour unauthorized access structurally offset by voluntary too-dangerous designation; sub-5pt net effect but material SYS/ACC negative on record.

03

South Sudan FINAL CYCLE: 7th consecutive deferral; UNMISS mandate expires April 30 (tomorrow); emergency human review mandatory April 30 morning — pipeline cannot defer again.

04

Floor-limitation methodology gap CRITICAL: 8 nights overdue; five entities (xAI 2.2, Palantir 6.6, Sudan 0, South Sudan 0, Israel 8.8) accumulating unexpressible evidence nightly.

05

Australia UN HRC review (January 26, 2026) and Singapore April 2026 OHCHR drug-execution moratorium call documented as first-ever rotation baselines for both countries.

06

OpenAI trial in progress (Day 3): Musk retook stand April 29; $150B potential damages; liability phase verdict approximately May 21; mandatory re-queue post-verdict.

The weekly briefing on institutional compassion scores

Score changes, sector trends, and emerging risk signals from overnight research across 1,155 entities — every Monday. Free.

No spam. Unsubscribe anytime. Your email is never shared.

Sector intelligence

Analyst-level observations on patterns emerging across indexed sectors from the Apr 29 research cycle.

AI Labs

  • Governance divergence entrenching across two cycles. Positive cluster (DeepMind 65.0, Anthropic 61.6) confirmed with proactive safety architecture. Negative cluster (xAI 2.2, Palantir 6.6) accumulating evidence at floor boundary. OpenAI (27.5) binary-pending trial outcome. Approximately 55-60 composite points separating the governance leaders from floor-limited entities — widest documented intra-index divergence in pipeline history. EU AI Act enforcement August 2 will function as external confirmation mechanism.

Countries — Conflict Zone

  • Sudan and South Sudan both at composite 0 with accelerating deterioration and nightly new evidence. Israel at 8.8 near-floor with famine confirmation and systematic aid access restriction. All three entities in floor-limitation cluster. Evidence quality and source credibility (UN bodies, OHCHR, OCHA, UNSC) has been consistently high across the cycle — the limitation is methodological, not evidential.

Fortune 500 — Technology

  • AI-labor displacement pattern confirmed across Meta, Microsoft, Amazon, Oracle (index gap) in April 2026. All profitable companies with no existential financial pressure; capital reallocation to AI infrastructure is the stated rationale. EMP dimension facing structural downward pressure across the technology sector. Voluntariness modifier needed before next full F500 rotation sweep.

US States and Cities

  • Texas-Houston cascade continues to compound. Texas (14.1 Critical) exhibiting floor pressure with EQU/ACC/INT dimensions constrained at 1.3 sub-band ceiling. Minnesota (84.4 Exemplary) confirmed stable despite federal enforcement activity within state borders, per attribution methodology. US immigration enforcement pattern documented across HRW, Amnesty, and AI reports represents a structural multi-state signal. Next cycle should assess additional Texas metros.

Emerging risks

Forward-looking risk signals from the Apr 29 research cycle. These are not current findings — they are early warning flags.

Risk

UNMISS mandate expiry April 30 (CRITICAL): if no renewal or technical rollover confirmed by April 30, South Sudan mandate lapse would represent unprecedented UN peacekeeping governance failure with direct civilian protection consequences. Hard deadline: check press.un.org and What's In Blue April 30 morning. 2025 precedent suggests technical rollover is likely but unconfirmed. Entities affected: south-sudan.

Risk

Musk v. Altman trial verdict (~May 21, HIGH): breach-of-charitable-trust claim is potentially precedent-setting for all AI nonprofits and PBCs. ~50B potential damages. Mandatory OpenAI re-queue post-remedy. Outcome could materially affect INT/SYS scoring depending on judicial findings on governance conversion. Entities affected: openai.

Risk

Anthropic Mythos breach investigation (HIGH): unauthorized access via guessable URL and vendor credential exploit within 14 hours of restricted launch. If investigation reveals systemic access-control failures, this becomes a standalone SYS/ACC downward signal in a future cycle. Currently offset but not neutralized. Entities affected: anthropic.

Risk

EU AI Act full enforcement August 2, 2026 (HIGH): 95 days to enforcement. Only 8 of required member-state contact points designated as of March 2026. Entities with ongoing regulatory conflicts — xAI (EU Commission DSA proceedings), Palantir (UK Parliamentary scrutiny) — face compounding exposure. Pre-enforcement guidance expected Q2 2026.

Risk

Floor-limitation methodology transparency gap (HIGH, 8 nights overdue): published scores for Sudan (0), South Sudan (0), xAI (2.2), Israel (8.8), and Palantir (6.6) increasingly understate the evidence weight documented by the pipeline. Benchmark credibility risk if gap grows without methodology-note resolution.

Risk

Texas floor-pressure acceleration (MEDIUM): composite 14.1 Critical with EQU/ACC/INT at 1.3 sub-band ceiling. Abbott fiscal coercion of cities, HRW Metro Surge documentation, Amnesty Texas-specific concerns. Next cycle evidence accumulation may push toward floor-limitation constraints. Entities affected: texas, houston.

Risk

AI labor cascade continuation (MEDIUM): 92,000+ tech layoffs YTD 2026; Meta, Microsoft, Amazon, Oracle all with AI-efficiency rationale. Pattern accelerating. Additional F500 entities not yet in T1 queue may announce comparable measures in Q2 2026. Pipeline EMP voluntariness modifier still undeveloped.

Risk

Oracle index gap (MEDIUM): March 31 30,000-employee layoff event unprocessable due to Oracle’s absence from live fortune-500.json. Every additional night widens the gap between scanner evidence and assessable evidence for one of the cycle’s most significant labor events.

Risk

Scanner baseline drift (LOW): Belgium (scanner 74.9 vs live 77.9), Singapore (73.3 vs 62.2), Portugal (68.6 vs 65.6), Japan (68.6 vs 65.6), Australia (65.5 vs 62.5), United States (35.5 vs 25.0). Divergences create priority-score distortions and could surface false proposals if uncorrected through next rotation cycle.

Research insights

Analytical observations from the Apr 29 research cycle. These are assessor-level interpretations, not findings.

Note

Second consecutive zero-proposal night confirms the pipeline's rotation-state drift gap has been substantively closed following batch application of all 16 proposals from April 25-27. The benchmark is now operating on a clean, evidence-anchored state for assessed entities — but this clean state masks a structural divergence at the floor boundary where 5 entities have accumulated material negative evidence the scoring system cannot express.

Note

The Anthropic Mythos breach is the most important new event from this assessment window that does NOT generate a score change. The SYS/ACC negative (14-hour unauthorized access via credential exploit) is materially offset by the positive AWR/INT signal from the voluntary too-dangerous-to-release classification itself — a structurally unusual pairing where the same model launch generates both a serious security failure and a meaningful safety-commitment signal simultaneously.

Note

The floor-limitation methodology gap has now accumulated across 8 nights without resolution. The five affected entities (Sudan 0, South Sudan 0, xAI 2.2, Israel 8.8, Palantir 6.6) collectively represent some of the most consequential institutional behaviors tracked by the benchmark. Every additional night without a methodology-note framework means published scores appear to represent the full state of evidence when they do not.

Note

South Sudan has reached its FINAL CYCLE before the April 30 UNMISS mandate expiry hard deadline. The 7th consecutive deferral cannot be repeated — April 30 morning human review is mandatory regardless of vote outcome. The 2025 precedent (technical rollover April 30, substantive renewal May 8) suggests the mandate will not formally lapse, but the pipeline cannot defer past the deadline without converting the deferral to a methodology-note registration.

Note

The rotation-state divergence between scanner baselines and live index data (6 entities flagged tonight: Belgium, Singapore, Portugal, Japan, Australia, United States) indicates the pipeline infrastructure needs a periodic baseline-sync mechanism. Every time the score-updater applies proposals to the live index without simultaneously updating rotation-state.json, the gap widens. This is a structural pipeline integrity issue distinct from any individual assessment quality question.

Note

Texas (14.1 Critical) is emerging as a floor-pressure candidate for the next assessment cycle. Abbott's documented coercion of Houston, HRW Operation Metro Surge documentation, and Amnesty Texas-specific concerns collectively represent a continuing downward evidence pattern. The sub-band EQU/ACC/INT positioning at 1.3 leaves minimal additional expressive room before floor-limitation constraints would apply to Texas as well.

Assessed entities

All entities assessed in tonight's research cycle, with composite scores and band classifications.

Want the complete picture?

Full benchmark reports include all 40 subdimension scores, complete evidence trails, and methodology documentation for every assessed entity.