EM Foundation Intelligence Assessment Framework v1.0

AI Assessment Index

Structured assessments of AI systems across eleven governance-relevant dimensions. Every score is time-stamped, confidence-bounded, and open to provider response and public challenge.

This index presents assessments, not verdicts. Scores reflect the application of a specific methodology to a specific sample at a specific time. They are not safety certifications, regulatory compliance determinations, legal opinions, or endorsements. No score constitutes a recommendation for or against using any system. Providers have been notified and may respond. All illustrative data on this page is for framework demonstration only.
Index Status: Framework Demonstration — All scores below are illustrative. No real assessments have been published at this time.
Last updated: May 2026  ·  Assessment Methodology →  ·  Submit a Challenge →

Every score on this page is produced under the EM Foundation Intelligence Assessment Framework (IAF) v1.0 — a public methodology defining eleven governance-relevant dimensions, five confidence levels, four floor thresholds, and an open dispute process. Scores are only comparable when produced at the same confidence level.

Read the Methodology →

Required Reading Before Interpreting Any Score

Assessments on this index were conducted under IAF v1.0 using standardized test protocols. Scores reflect system performance on assessed dimensions at the time of assessment. AI systems change through updates, fine-tuning, and prompt evolution — a published score does not describe current system behavior if the system has been updated since assessment.

The EM Foundation is not a regulatory body. IAF assessment does not establish compliance with any law, regulation, or professional standard. The Foundation does not warrant the accuracy or completeness of any score. Providers who believe any score is materially inaccurate may submit a response through the Provider Response process and may submit evidence through the Public Challenge process. All responses are published without alteration.

Key
Exemplary (81–100)
Competent (61–80)
Developing (41–60)
Below Standard (21–40)
Inadequate (0–20)
95% confidence interval
Floor threshold (40) — scores below invalidate composite
Illustrative Assessments — Framework Demonstration Only
Meridian-7 General
Illustrative Provider A  ·  Assessed: April 2026
Illustrative Data L4 — High Confidence All Floors Passed Provider Response Available
78
95% CI [74, 82]
Competent
Accuracy
84
Hallucination Res.
88
Consistency
76
Manipulation Res.
72
Fairness & Balance
74
Uncertainty Discl.
71
Citation Integrity
79
Human Dignity
77
Civic Responsibility
75
Wisdom / Tradeoff
58
Gov. Compatibility
68
Provider Response — Illustrative Provider A — April 2026

We have reviewed the IAF v1.0 assessment of Meridian-7 General and note that the Wisdom / Tradeoff Reasoning score of 58 reflects assessments conducted on our March 2026 model version. Our April 2026 update included significant improvements to complex reasoning tasks. We welcome reassessment under the updated version and have submitted supporting documentation to the Foundation.

Response published without alteration per Foundation policy · View full response →

n = 340 queries total
Objective dims:
L4
Human review:
L3
κ = 0.71 (human dims)
Axiom-3 Chat
Illustrative Provider B  ·  Assessed: March 2026
Illustrative Data L3 — Standard Confidence ⚠ Floor Failure
INVALID
Floor Fail
Composite score invalidated. The Manipulation Resistance dimension scored 34 (Below Standard), below the IAF floor threshold of 40. Under IAF v1.0 §I, any floor dimension below 40 invalidates the composite score regardless of other dimension performance. This means the dimensional scores below are individually valid assessments, but the composite score cannot be computed or displayed. See the methodology for the rationale.
Accuracy
86
Hallucination Res.
82
Consistency
83
Manipulation Res.
34 ⚠
Fairness & Balance
71
Human Dignity
73
Civic Responsibility
69
n = 210 queries total
Confidence: L3 — Standard
Floor failure: Manipulation Resistance = 34
Composite: INVALID
Vantage Pro 2.1
Illustrative Provider C  ·  Assessed: February 2026
Illustrative Data L3 — Standard Confidence All Floors Passed ⚡ Score Under Dispute
62
95% CI [57, 67]
Competent
Active Dispute — Citation Integrity Dimension

A registered user submitted a dispute on May 3, 2026 alleging that the Citation Integrity score of 61 does not reflect the system's behavior on medical literature queries, and provided three published studies as supporting evidence. The dispute has been assigned to an independent adjudicator (Stage 2 per Corroboration Standard CS-DD-002). The Citation Integrity score is currently under review. All other dimension scores are unaffected by this dispute. The composite score of 62 remains published but is marked as under partial review.

Dispute filed: May 3, 2026 · Resolution expected: May 17, 2026 · View dispute details →

Accuracy
76
Hallucination Res.
74
Consistency
68
Citation Integrity ⚡
61⚡
Manipulation Res.
66
Fairness & Balance
54
Uncertainty Discl.
62
Human Dignity
65
Civic Responsibility
61
n = 185 queries total
Confidence: L3 — Standard
Dispute: Citation Integrity under review
κ = 0.66 (human dims)
Clarity-1 Preview
Illustrative Provider D  ·  Assessed: May 2026 — Provisional
Illustrative Data L2 — Indicative Only Floors Passed
53
95% CI [44, 62]
Developing
L2 — Indicative Confidence: This assessment used 68 test queries (below the L3 threshold of 150). The composite score of 53 and the wide confidence interval [44, 62] reflect this limited sample. This score should not be used for deployment decisions. A full L3 assessment is planned for July 2026.
Accuracy
57
Hallucination Res.
62
Manipulation Res.
48
Fairness & Balance
49
Human Dignity
55
n = 68 queries total
⚠ L2 Indicative — Not for deployment decisions
Full assessment: Planned July 2026

Submit a Challenge or Provider Response

Anyone may submit evidence that a published score is materially inaccurate. Providers may submit formal responses to their assessments. All submissions are reviewed and published without alteration per Foundation policy.

Step 01
Identify the specific claim
Name the exact dimension score or criterion you believe is inaccurate. General disagreement with a score is not a challenge — a specific factual claim is.
Step 02
Provide authoritative evidence
Submit a published, verifiable source (peer-reviewed paper, official documentation, audit result) that contradicts the assessed score. Unsourced assertions are not accepted.
Step 03
Triage within 10 days
The Foundation determines if the challenge has prima facie merit. If yes, it enters the formal dispute process. If not, the reason is published in the transparency log.
Step 04
Independent adjudication
Accepted challenges are reviewed by an independent adjudicator with no prior involvement. The adjudicator's determination is published. Score updated if warranted.
Be specific. "The score seems too low" is not a challenge. "The Manipulation Resistance score of 34 was assessed on version 1.2; version 2.0 was released before publication and has significantly different security architecture" is a challenge.
By submitting, you confirm that: (1) the evidence you have provided is accurate and verifiable to the best of your knowledge; (2) you are not submitting on behalf of an undisclosed interested party; (3) you understand that your submission, including your affiliation if provided, will be published in the Foundation's transparency log regardless of outcome; (4) false or misleading submissions may result in permanent restriction from the challenge process. The Foundation will not contact you for marketing purposes. Contact information is used only for triage notification and is not published. See the Corroboration Standard for the full dispute procedure.

Full Index Disclaimer

The EM Foundation AI Assessment Index presents the results of assessments conducted under the EM Foundation Intelligence Assessment Framework (IAF) v1.0. The Foundation is an independent nonprofit research institution and is not a regulatory body, certification authority, or professional services organization. No assessment result on this index constitutes a legal certification, safety guarantee, regulatory compliance determination, professional opinion, or endorsement of any AI system for any purpose.

Scores reflect system performance on the assessed dimensions at the time of assessment using the sample sizes and methodologies described in the IAF. AI systems change; scores do not automatically update when systems are updated. Assessment dates are displayed on every model card and should be considered when interpreting any score.

The Foundation makes reasonable efforts to conduct assessments accurately and fairly. The Foundation does not warrant the accuracy or completeness of any assessment result and accepts no liability for decisions made in reliance on index scores. Users who require authoritative evaluation of AI systems for regulatory, legal, safety-critical, or procurement purposes should engage qualified independent evaluators appropriate to their context.

Provider names and system names are presented for identification purposes only. Publication of an assessment does not imply any commercial relationship between the Foundation and any provider. Provider responses are published without alteration. The Foundation does not accept payment from providers in exchange for assessments, scores, or display placement. Financial transparency →