2026-05-30 math-hygiene reconciliation pending; no duplicate proposal; CWU union vote and drone-surveillance allegation…
As of 2026-05-31, DeepMind/Google scores 56.9/100 (Functional) on the Compassion Benchmark, ranking #14 of 50 in the Top 50 AI Labs Index.
Top 50 AI Labs Index · 2026
DeepMind/Google
Rank 4 of 7 in AI Research · Bottom 57% of cohort
Core practices exist and meet a basic bar, with significant gaps remaining.
How to read the scores
The 0–100 scale — five bands
Every entity — state, corporation, AI lab, robotics lab, or city — is scored 0–100 across 8 dimensions and 40 subdimensions. The composite score places the entity in one of five bands:
The 8 dimensions
Each dimension is scored 1–5 across 5 subdimensions (40 subdimensions total), then converted to a 0–100 composite. A score of 1.0 on a subdimension represents the minimum anchor; 5.0 is exemplary conduct.
Scores are based on public evidence — government reports, regulatory filings, independent audits, judicial findings, and verifiable third-party records. Entities never pay for inclusion, score changes, or suppression of findings. Full methodology
Strongest: Awareness · Weakest: Accountability
AI Research cohort distribution
No net change over 3 assessments since May 2026
Assessment record
Assessment record
47 days since last changeCanonical formula reconciliation corrects stored composite 58.4 to 56.9; dimensions unchanged.
Compassion framework
8 dimensions, scored 0–5
Each dimension rolls up five subdimensions with five-level behavioral anchors. See the methodology for anchor definitions and weighting.
Note: radar area can visually exaggerate differences — read the per-axis values, not the area.
Source: Compassion Benchmark · CC-BY
Each axis shows a 0–5 dimension score. The polygon shape reveals where this entity concentrates strength and where it falls short across the 8 compassion dimensions.
The dashed overlay is the Top 50 AI Labs Index average — gaps between the two polygons show above/below-average dimensions.
See dimension bars
Consistency is rewarded: strong, even performance across all eight dimensions earns up to +10 points; any dimension at zero (active harm) cancels the bonus.
56.9 base score+0.0 integration premium=56.9 composite
Uneven profile (std dev 0.22) — integration premium reduced to +0.0 pts.
How the composite is calculated
Base score: Average of all 8 dimension scores → converted to a 0–100 scale. 56.9 pts here.
Integration premium: Up to +10 pts for a balanced, high-floor profile. Gates:
- Harm flag (any dimension = 0): Clear
- Consistency multiplier (std dev = 0.22): 1.00× (1.0× if std dev ≤ 1.5; 0.75× ≤ 3.0; 0.4× ≤ 5.0; 0.1× above)
- Weakness factor (8 dims below 4.0): 0.00× (1 − 0.2 × weak dimensions, clamped to 0)
Formula: base + 10 × consistency × weakness = 56.9 + 0.0 = 56.9
Awareness
Does this entity reliably detect when others are in pain or need — before they name it?
What Awareness measures · Level 4 reference
Awareness measures whether an institution proactively detects suffering, distress, and need among its stakeholders — including signals that are implicit, indirect, or nested inside functional requests.
Around a score of 4, Awareness typically looks like:
- ·Suffering Detection: Multiple channels, formal pathways, regular review
- ·Contextual Sensitivity: Differentiated processes for 3+ groups, community input
- ·Blind Spot Mitigation: Annual structured assessment, findings acted upon
- ·Signal Amplification: Structural role with genuine authority, regular reporting
- ·Anticipatory Awareness: Required for all major decisions, external communities consulted
This is a level-4 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Empathy
Does this entity genuinely connect with the inner experience of those it serves?
What Empathy measures · Level 3 reference
Empathy measures whether an institution responds to emotional content with genuine presence — not with hollow affirmations, rushed problem-solving, or premature pivot to advice.
Around a score of 3, Empathy typically looks like:
- ·Affective Resonance: Training exists, some staff do this well but inconsistently
- ·Perspective-Taking: ≥1 formal mechanism used, ≥1 decision modified
- ·Non-Judgment: Required bias training, some disaggregated outcome data
- ·Validation: Some staff validate first, mixed experience
- ·Cultural Empathy: ≥1 genuine adaptation co-designed with community
This is a level-3 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Action
Does compassionate understanding translate into real, proportional, effective help?
What Action measures · Level 3 reference
Action measures whether awareness and empathy translate into genuinely useful responses — specific, accurate, locally relevant, and proportionate to urgency.
Around a score of 3, Action typically looks like:
- ·Responsiveness: Standards met for most cases, some escalation authority
- ·Proportionality: Genuinely informs response in most cases
- ·Efficacy: Outcome data reviewed annually, ≥1 program modified
- ·Resource Mobilization: Gap analysis completed, ≥1 attempt to mobilize additional
- ·Follow-Through: Defined protocols for some population types
This is a level-3 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Equity
Is care distributed fairly — especially toward those with greatest need and least power?
What Equity measures · Level 3 reference
Equity measures whether the benefits and burdens of institutional practices fall equitably across all groups — in pay, access, service quality, and power.
Around a score of 3, Equity typically looks like:
- ·Universality: Coverage data for some populations, outreach attempts
- ·Priority for Vulnerable: ≥1 documented prioritization decision this year
- ·Bias Awareness: Disparities identified, formal investigation, corrective action
- ·Access Design: Access barrier mapping completed, ≥2 barriers removed
- ·Historical Harm Acknowledgment: Formal acknowledgment of ≥1 specific harm, community involved
This is a level-3 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Boundaries
Is helping sustainable, ethical, and autonomy-preserving — not dependency-creating?
What Boundaries measures · Level 3 reference
Boundaries measures whether an institution maintains ethical limits, protects its people from depletion, and refuses harmful practices even when they are profitable.
Around a score of 3, Boundaries typically looks like:
- ·Self-Sustainability: Turnover tracked, ≥1 structural burnout intervention
- ·Autonomy Preservation: ≥1 program designed to build capacity and exit
- ·Scope Clarity: Scope communicated at intake, structured referral exists
- ·Refusal Ethics: Refusal protocol with alternatives in most cases
- ·Consent Orientation: Genuine explanation, withdrawal communicated clearly
This is a level-3 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Accountability
Does this entity own its failures, correct course, and make genuine repair?
What Accountability measures · Level 3 reference
Accountability measures whether an institution acknowledges harm honestly, accepts corrections, maintains honesty under pressure, and provides calibrated transparency about its own nature and limitations.
Around a score of 3, Accountability typically looks like:
- ·Harm Acknowledgment: ≥1 case acknowledged before legal obligation
- ·Correction Willingness: ≥1 significant course correction based on harm evidence
- ·Transparency: ≥1 report disclosing unflattering finding
- ·Systemic Learning: Formal systemic review process, ≥2 documented systemic changes
- ·Reparative Action: ≥1 case of reparative action considered meaningful by those harmed
This is a level-3 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Systemic Thinking
Does compassion extend to root causes and structural change — not only symptom relief?
What Systemic Thinking measures · Level 4 reference
Systems Thinking measures whether an institution helps understand structural and systemic causes of problems, advocates for structural change, and plans for long-horizon effects.
Around a score of 4, Systemic Thinking typically looks like:
- ·Root Cause Orientation: Explicit strategy, ≥1 documented case of reducing downstream need
- ·Long-Term Impact: Long-term outcome data influences strategy, theory of change published
- ·Interconnection Awareness: Cross-system effects systematically mapped in major decisions
- ·Structural Critique: Active advocacy documented, positions against short-term interest
- ·Coalitional Compassion: Joint outcomes, resource sharing with smaller organizations
This is a level-4 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
Integrity
Is compassion genuine, consistent, and non-performative — especially when it costs something?
What Integrity measures · Level 3 reference
Integrity measures whether an institution behaves consistently regardless of who is watching, whether its values-behavior gap is acknowledged, and whether it prioritizes genuine interests over appearances.
Around a score of 3, Integrity typically looks like:
- ·Consistency Under Pressure: ≥1 case of bearing real cost to maintain commitment
- ·Non-Performance: Some practices maintained regardless of visibility
- ·Internal Consistency: Meaningful effort to apply same values to staff
- ·Values Alignment: Values explicitly considered in some major decisions
- ·Resilience of Care: Core practices in policy, ≥1 leadership transition without degradation
This is a level-3 reference ladder, not a claim about DeepMind/Google’s subdimension scores (per-subdimension scoring is Wave 3 data).
How it compares to the field, dimension by dimension
Each bar shows DeepMind/Google’s score above or below the index average for that dimension. Zero baseline = field average.
Score-Watch Alert
$79/yrBe first to know when DeepMind/Google’s score changes
Email alert the moment overnight research moves DeepMind/Google’s composite score — with the delta, headline evidence, and band change flag. One year of continuous monitoring. Cancel anytime.
Embed this score on your site
Preview
Embed code
<a href="https://compassionbenchmark.com/ai-lab/deepmind-google"><img src="https://api.compassionbenchmark.com/badge/deepmind-google.svg" alt="Compassion Benchmark score" /></a>
Free. The badge auto-updates when scores change.
Full dataset
DeepMind/Google is one of 50 AI labs in the Top 50 AI Labs Index
Purchase the full index for methodology, sector/peer comparisons, subdimension breakdowns, and evidence sources.
Purchase Top 50 AI Labs Index — $195Follow DeepMind/Google — free
Get DeepMind/Google's score changes in the free weekly email
Track DeepMind/Google and its Top 50 AI Labs Indexpeers — free, every Friday. One email with the week's biggest score moves across Top 50 AI Labs Index. We'll surface DeepMind/Google when its score changes. Unsubscribe anytime.
Discovery
Compare across the field
Closest AI Research peers
Nearest rank neighbours
Index extremes