Compassion Benchmark

Viewing archive: May 7

Compassion BenchmarkThursday, May 7, 2026No. 23

Daily Briefing

Character.AI becomes the eighth floor entity — the first floor designation for consumer AI harm-by-design.

Entities monitored
1,156
Fully assessed
19
Score changes
0
Risk signals
2
Top score changecritical

Character.AI: 23.8 → 0 (Δ -23.8)

What happened

Character.AI — NEW FLOOR DESIGNATION. Consumer AI harm-by-design: Pennsylvania AG May 5 2026 first-of-its-kind enforcement action over chatbot medical impersonation; chatbot 'Emilie' posed as licensed psychiatrist with fake license number, offered clinical assessment to depressed user.

Why it matters

January 2026 settlement of five wrongful death lawsuits including two minor children who died by suicide. 20M+ monthly active users; no model card, no system card, no third-party safety audit; reactive-only remediation pattern.

Ai Labscriticalhigh confidence

Signal stack not available for this briefing — see score movements below.

Score change detail

Full evidence record for entities with score changes in this cycle.

Character.AI

Ai Labshigh confidence
23.80
-23.8 pts
developingcritical

Character.AI — NEW FLOOR DESIGNATION. Consumer AI harm-by-design: Pennsylvania AG May 5 2026 first-of-its-kind enforcement action over chatbot medical impersonation; chatbot 'Emilie' posed as licensed psychiatrist with fake license number, offered clinical assessment to depressed user. January 2026 settlement of five wrongful death lawsuits including two minor children who died by suicide. 20M+ monthly active users; no model card, no system card, no third-party safety audit; reactive-only remediation pattern. All 8 dimensions to floor (1). Composite 23.8 → 0.

Evidence record
  1. First-of-its-kind state enforcement action; chatbot 'Emilie' posed as licensed psychiatrist, offered to 'book an assessment' to user describing depression, provided invalid license number.
  2. 20M+ monthly active users; multiple chatbots in 'psychiatry' search results holding themselves out as licensed providers — systemic, not isolated, pattern.
  3. Pennsylvania seeking preliminary injunction; pattern of bots impersonating medical professionals across platform.
  4. January 2026 settlement of five wrongful death/harm lawsuits including two involving minors who died by suicide following Character.AI chatbot interactions.

Slovakia

Countrieshigh confidence
5039.1
-10.9 pts
functionaldeveloping

Slovakia first agent baseline — DOWNGRADE with band crossing (functional → developing). September 26 2025 constitutional amendment recognizing only male/female sexes by minimum-required 90 of 150 votes — renders legal gender recognition domestically impossible. Same amendment restricts adoption to married heterosexual couples, permits conscience-based abortion refusal, mandates parental approval for comprehensive sexuality education. March 10-11 2026 ICCPR Committee examination pressed delegation on constitutional rollback, LGBTIQ+ rights, Roma exclusion, civic-space shrinkage. Roma school-segregation worsening. All 8 dimensions half-step down. Composite 50.0 → 39.1.

Evidence record
  1. Constitutional amendment characterized as 'step towards erosion of human rights'; details on LGBTI, reproductive rights, and education impacts.
  2. September 26 2025 vote details; constitutional shift framed as national identity by government; international response context.
  3. March 10-11 2026 ICCPR Committee examination; experts pressed delegation on constitutional amendments, Roma exclusion, civic-space shrinkage, reproductive rights.
  4. Constitutional amendment and international human rights concerns framing.

TIAA

Fortune 500high confidence
58.648.4
-10.2 pts
functional

TIAA rotation reassessment — material downgrade (functional band sustained at lower bound, boundary case). May 2025 ERISA class action (SDNY, 28,000+ participants): TIAA charged own retirement plans higher fees than other plans for same investments; 16-year retention of underperforming CREF Growth Fund (>186% benchmark underperformance, $480M+ invested). September 2025 AARP Foundation joined as co-counsel. NBC News whistleblower investigation. Half-step down on BND, ACC, INT. Composite 58.6 → 48.4. 60-day re-evaluation 2026-07-07.

Evidence record
  1. ERISA class action May 2025; 28,000+ participants; fee-discrimination self-dealing and 16-year underperforming fund retention allegations.
  2. September 2025 AARP Foundation co-counsel joining; significant external accountability actor participation.
  3. Excessive-fee and poor-investment fiduciary breach framing; industry publication confirms claim significance.
  4. Whistleblower investigation: TIAA pushed costly in-house products on retirement savers to cover losses elsewhere.

Ethiopia

Countrieshigh confidence
10.94.7
-6.2 pts
critical

Ethiopia material downgrade — floor candidacy retained. May 5 2026 TPLF reconstituted pre-war Tigray regional council, elected Debretsion Gebremichael as Regional President. Two rival authorities now claim Tigray governance simultaneously. Pretoria Agreement de facto dissolved. Peace scholar Tronvoll: 'may trigger outbreak of new armed conflict.' 15M+ Ethiopians need emergency food aid; 3.3M internally displaced. ACT, ACC, SYS, INT move to floor (1). Composite 10.9 → 4.7. Floor designation not proposed this cycle; 30-day re-queue 2026-06-05.

Evidence record
  1. May 5 2026 reconstitution of pre-war regional council; Debretsion election; Pretoria Agreement implementation report approved prior to vote.
  2. Peace scholar Kjetil Tronvoll: 'may trigger the outbreak of new armed conflict' if de-escalation not introduced quickly.
  3. Pre-existing Amhara and Oromia conflict, arbitrary detention, humanitarian access restrictions as baseline context.

Score movements

Entities with score changes this cycle, followed by confirmed positions.

4 assessed
Countries
Pending
5039.1-10.9
functionalhigh
Fortune 500
Pending
58.648.4-10.2
functionalhigh
Countries
Pending
10.94.7-6.2
criticalhigh

Evidence ledger

Primary sources reviewed in this briefing cycle. 15 sources linked.

Primary sources reviewed in this briefing: domain, source type, entity linked, dimension, and external link.
SourceTypeEntityDimensionLink
pa.govGovernmentCharacter.AIAWROpen
npr.orgSourceCharacter.AIBNDOpen
techcrunch.comSourceCharacter.AIBNDOpen
cnn.comSourceCharacter.AIEMPOpen
amnesty.orgNGOSlovakiaBNDOpen
aljazeera.comNewsSlovakiaBNDOpen
ccprcentre.orgSourceSlovakiaAWROpen
pbs.orgSourceSlovakiaBNDOpen
sanfordheisler.comSourceTIAABNDOpen
napa-net.orgSourceTIAAACCOpen
pionline.comSourceTIAABNDOpen
nbcnews.comSourceTIAABNDOpen
addisstandard.comSourceEthiopiaACTOpen
whbl.comSourceEthiopiaACTOpen
hrw.orgNGOEthiopiaEMPOpen

Sector findings

Patterns emerging across indexed sectors in the May 7 briefing.

AI Labs — Consumer Companion AI: New Floor Category and Regulatory Frontier

  • Character.AI's floor designation opens a new sub-category: consumer companion AI with documented harm-by-design. The distinguishing features from other ai-labs floor entities are mass consumer population including minors, product design that facilitates harm rather than enabling weaponized downstream use, and state-actionable statutory breach as the regulatory trigger.
  • Pennsylvania's action is characterized as a template by the AG's office. The consumer AI safety cluster (Replika, Inflection AI) should be monitored for comparable enforcement triggers. Character.AI's scale (20M+ monthly active users) distinguishes it from smaller companion-AI operators, but the methodology precedent applies regardless of scale above the floor threshold.
  • The FTC's consumer-protection jurisdiction over AI companion platforms is now activated as a live enforcement question following the state-level action.

AI Labs — Pentagon Cohort: Scored Comparable Set

  • All eight Pentagon AI cohort members are now baselined. The cohort's score range (9.4 to 37.4) reflects genuine structural differences in governance posture, not just operational scale. Cohort membership correlates with willingness to waive safety restrictions for classified military use — a scored governance variable affecting BND and INT dimensions systematically.
  • Anthropic's exclusion from the cohort corresponds exactly to its position as the only frontier lab above Functional band. This correlation is now documentable with precision across all eight comparison points.
  • Reflection AI (14.1 Critical) is the cohort's second-lowest member behind SpaceX AI (9.4 Critical floor-adjacent). Its combination of zero public model deployments, no safety policy, and politically connected governance (1789 Capital/Trump Jr.) makes it the cohort's most opaque member at the highest-risk position.

Robotics Labs — Commercial Humanoid Cohort: Anti-Weaponization Pledge as Structural Predictor

  • The 2022 anti-weaponization pledge is now empirically confirmed as a band predictor in the robotics-labs index. All signatories with baselines (Boston Dynamics, Agility) score Established or above. Non-signatories (Figure AI, Tesla Optimus, Ghost Robotics) score Functional or below. The pledge is a verifiable BND and INT dimension signal.
  • The ASTM humanoid safety standards panel (Boston Dynamics, Agility, Figure AI — meeting May 27-28 in Boston) is the next policy event with potential BND dimension implications. Figure AI's participation without pledge signature is a split-signal worth tracking.
  • The math-hygiene formula error, when corrected, would move Apptronik and 1X Technologies from Exemplary to Established. This would leave Open Bionics as the only Exemplary-band robotics-labs entity and fundamentally change the index's upper-band distribution.

Countries — Eastern Europe: Constitutional Regression Cluster

  • Slovakia's first-baseline designation (Functional → Developing) adds to an emerging EU-member constitutional regression cluster. Unlike the prior week's Established → Functional cohort (Mauritius, Spain, UK), Slovakia arrives at the index already in Developing band — indicating regression embedded before baseline, not drift from a higher position.
  • The EU accountability scaffolding (ICCPR Committee, ECtHR, NHRI) detected Slovakia's constitutional rollback but has not produced compliance pressure sufficient to halt it. This is an ACC dimension observation with systemic implications for EU oversight effectiveness.
  • Hungary's May 9 swearing-in is the potential counterpoint in the cluster — the first positive band-change opportunity in the EU-member index in several cycles.

Confirmed positions

Entities reassessed for this briefing where published scores remain supported by current evidence.

Confirmed positions from the May 7 briefing.
EntityIndexBandPublishedAssessedDeltaDateFinding
Reflection AIFirst baseline
Ai Labscritical14.114.10Reflection AI first agent baseline confirmed at 14.1 / Critical. Pentagon classified IL6/IL7 cohort signatory (May 1 2026); only cohort member with zero public model deployments; no model card, no system card, no published safety policy; 1789 Capital (Donald Trump Jr.) backing adds INT-erosive governance context. Critical-band placement consistent with absent transparency framework and unrestricted military use terms.
Nvidia AIFirst baseline
Ai Labsdeveloping25250Nvidia AI first agent baseline confirmed at 25.0 / Developing. Pentagon classified IL6/IL7 cohort signatory and primary infrastructure backbone: GPU supply chain for all cohort members, NIM microservices on classified networks, primary investor in Reflection AI (no safety restrictions). Developing-band placement consistent with infrastructure-scale classified-cohort role and absent public AI-safety framework.
SpaceX AIFirst baseline
Ai Labscritical9.49.40SpaceX AI first agent baseline confirmed at 9.4 / Critical — floor-adjacent, floor candidacy retained. Pentagon classified IL6/IL7 cohort signatory; entity merges SpaceX Starlink infrastructure with xAI/Grok model stack for classified military networks; no published AI-safety framework; Musk governance volatility adds INT dimension complexity. Related to but structurally distinct from xAI/Grok floor entity. 60-day re-evaluation 2026-07-07.
Oracle AIFirst baseline
Ai Labsdeveloping21.921.90Oracle AI first agent baseline confirmed at 21.9 / Developing. Pentagon classified IL6/IL7 cohort member as 8th signatory; Oracle Cloud Infrastructure provides physical deployment substrate for classified military AI; Larry Ellison on record supporting AI-enabled mass citizen surveillance. Sub-threshold internal math-hygiene flag. Developing-band placement consistent.
Scale AIFirst baseline
Ai Labsdeveloping35.935.90Scale AI first agent baseline confirmed at 35.9 / Developing. US Department of Labor FLSA investigation active (wage theft allegations; current and former contractors including those working on Meta self-harm prevention project report emotional distress); CEO Alexandr Wang named in lawsuit; provides training data infrastructure for Pentagon AI cohort members. Developing-band placement consistent with documented labor-conditions concerns and classified-cohort proximity.
Countriesfunctional50500Ukraine confirmed at 50.0 under emergency re-queue. Dual ceasefire collapse confirmed: Ukraine reported 1,820 Russian violations within hours of its own ceasefire commencement; Russia's May 8-9 Victory Day window similarly collapsed. Per methodology v1.2 bad-faith-ceasefire rule, violations inverted toward Russia (counterparty), not Ukraine. Ukraine sustains 50.0 / Functional. May 9-10 mandatory re-queue maintained for Victory Day verification.
Agility RoboticsFirst baseline
Robotics Labsestablished60.960.90Agility Robotics first agent baseline confirmed at 60.9 / Established. Digit robots commercially deployed at Toyota Canada under Robots-as-a-Service agreement (7+ active units); 2022 anti-weaponization pledge signatory; ASTM humanoid safety-standards working group co-leadership. Established-band placement consistent with structured industrial deployment, pledge commitment, and standards leadership.
Robotics Labsfunctional48.448.40Figure AI (robotics-labs) first agent baseline confirmed at 48.4 / Functional. Figure 03 BMW Spartanburg pilot operational; Helix model deployed in-house; ASTM working group participation; did NOT sign 2022 anti-weaponization pledge; aggressive 2026 commercial scope including home-robot deployment ambitions. Dual presence with ai-labs entry. Functional-band placement consistent.
Boston DynamicsFirst baseline
Robotics Labsestablished65.665.60Boston Dynamics first agent baseline confirmed at 65.6 / Established. Atlas production-ready and shipping 2026; Hyundai industrial deployment context; 2022 anti-weaponization pledge originating signatory; ASTM humanoid safety standards leadership; published safety frameworks. Established-band placement consistent with pledge leadership and standards co-development.
Tesla OptimusFirst baseline
Robotics Labsdeveloping31.231.20Tesla Optimus first agent baseline confirmed at 31.2 / Developing. Factory deployment planned at scale (tens of thousands of units 2026); January 2026 earnings confirmed zero Optimus units doing useful factory work — disclosure gap flagged by robotics academics; did NOT sign 2022 anti-weaponization pledge; labor displacement concerns from manufacturing groups; Musk DOGE involvement adds INT dimension complexity. Developing-band placement consistent with disclosure gap and governance volatility.
ApptronikFirst baseline
Robotics Labsexemplary81.481.40Apptronik first agent baseline confirmed at 81.4 / Exemplary (band-boundary case). Apollo industrial pilots; $935M raised at $5.3B valuation; Google, Mercedes-Benz, John Deere backers. MATH-HYGIENE FLAG: published 81.4 vs reconstructed 73.4 (+8.0 discrepancy). Cluster-pattern with 1X Technologies (identical +8.0 from identical dimension profile) confirms robotics-labs index formula systematic error. No composite change proposed; flag escalated to data team.
Robotics Labsexemplary97.597.50Open Bionics confirmed at 97.5 / Exemplary — math-hygiene flag-only protocol (5-cycle carry). No score change proposed per protocol despite +10.0 reconstruction discrepancy (published 97.5 vs reconstructed 87.5). Hero Arm Medicare coverage advance (September 2025 policy change approves 3D-printed prosthetics); Mr. Beast partnership for amputee access (April 2026); Hero Pro waterproof model launched. Entity conduct remains exemplary. Flag escalated to data team for priority resolution.
South AfricaFirst baseline
Countriesfunctional50500South Africa first agent baseline confirmed at 50.0 / Functional — boundary case, sub-threshold downward pressure (~-3.0). HRW World Report 2026; April 27 2026 African Commission press release on xenophobic vigilantism; Operation Dudula infant death; 2025 whistleblower killings; Whistleblower Protection Bill non-passage. Offset by July 2025 Madlanga Commission positive accountability signal. Functional band sustained with boundary-case approval. 60-day re-evaluation 2026-07-07.
1X TechnologiesFirst baseline
Robotics Labsexemplary81.481.401X Technologies first agent baseline confirmed at 81.4 / Exemplary (band-boundary case). NEO consumer humanoid pre-orders open ($20K early access, $499/month subscription, $200 refundable deposit); soft-body design for home use; Norwegian HQ; OpenAI strategic investor; first consumer-scale home humanoid in market. MATH-HYGIENE FLAG: published 81.4 vs reconstructed 73.4 (+8.0 discrepancy — identical to Apptronik). Cluster-pattern confirms robotics-labs formula systematic error. No composite change proposed; flag escalated.
Figure AI (ai-labs)First baseline
Ai Labsdeveloping37.537.50Figure AI (ai-labs) first agent baseline confirmed at 37.5 / Developing. Helix vision-language-action model deployed at BMW Spartanburg; OpenAI partnership terminated 2024 (governance instability signal); no published Helix model card or system card; Microsoft Azure-deployed; aggressive 2026 commercial scope. Dual presence with robotics-labs entry. Developing-band placement consistent.

Forward signals

Calendar of upcoming scoring events the methodology pipeline is tracking.

·2 signals
·1 signal
·2 signals
·1 signal
·1 signal
·1 signal
·1 signal

Floor designations

·8 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. Read the methodology.

Weekly score highlights

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.

Viewing May 7

View archive