TruaceTracing the truth around AIFriday, August 28, 2026
The Index

What the evidence says.What the public feels.

Ranks distinct AI gain and problem claims from the published record. Scores reward impact, independent source strength, scale, confidence, and recency.

1,182 results
Show filters and sorting

AI gains · 658

68
GainHealth· Stable· Evidence: Moderate (1 source)

Machine learning models integrating multimodal big data improved precision transfusion management for AML patients by predicting transfusion demand and assessing transfusion reaction risks.

This peer-reviewed review published July 24, 2026 synthesized recent progress on machine learning models that integrate multimodal big data such as electronic health records, genomic and proteomic data to guide transfusion support for acute myeloid leukaemia, a highly heterogeneous malignancy where transfusion is essential.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Jul 27, 2026 · TRV-2026-0577

68
GainHealth· Stable· Evidence: Moderate (1 source)

An XGBoost model trained on routinely collected prehospital vitals and demographics reduced under-triage and over-triage compared to a surrogate of the 2021 National Field Triage Guidelines in 1.2M trauma patients.

Using a national trauma databank of over 1.2 million ambulance-transported adults from 2017-2020, researchers built an XGBoost model from routinely collected prehospital vitals, demographics, and field triage criteria to predict serious injury and compared it to a database-derived surrogate of the 2021 National Field Triage Guidelines.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Jul 27, 2026 · TRV-2026-0576

68
GainHealth· Stable· Evidence: Moderate (1 source)

Clinicians assisted by Retina4IRD achieved higher top-5 genetic diagnosis accuracy and better downstream management scores for suspected inherited retinal disease compared to specialist-only care.

Researchers developed Retina4IRD, an AI-based clinician decision support system that predicts genotype categories from fundus photographs and OCT scans, and tested it in internal and external validation and in a 300-participant randomized controlled trial comparing AI-assisted specialists to specialists alone for suspected inherited retinal diseases.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Jul 27, 2026 · TRV-2026-0575

68
GainHealth· Stable· Evidence: Moderate (1 source)

Random Forest and Gradient Boosting models trained on routinely collected clinical records predicted Hoehn and Yahr scores 5 years after ioflupane SPECT imaging, with best accuracy using 2 years of follow-up data.

Researchers developed and validated Random Forest and Gradient Boosting models to predict Hoehn and Yahr scores 5 years after 123I-ioflupane SPECT imaging, using harmonized data from 343 real-world patients and 134 PPMI patients with 83 overlapping features. Models using 2 years of clinical follow-up achieved the highest accuracy, driven by early H&Y scores, gait severity, and select imaging features.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%94

Updated Jul 27, 2026 · TRV-2026-0574

AI problems · 524

68
ProblemHealth· Stable· Evidence: Moderate (1 source)

AI struggles to generate novel hypotheses and apply creative, contextual reasoning, limiting its utility in rare disease diagnosis and complex uncertain scenarios.

A 2026 narrative review and conceptual analysis in Diagnosis synthesized literature on noise in medical decision-making, AI applications in healthcare, and clinical reasoning, reviewing case studies in radiology and pathology and empirical data on AI performance.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%91

Updated Jul 14, 2026 · TRV-2026-0218

68
ProblemScience· Stable· Evidence: Moderate (1 source)

Deep neural networks are unable to learn multiple tasks sequentially, suffering catastrophic forgetting when trained on new tasks.

On March 14, 2017, authors in PNAS described a method to overcome catastrophic forgetting in deep neural networks. They noted that while deep networks were the most successful technique for translation, image classification and generation, they could not learn multiple tasks sequentially. Their solution protects weights important for previous tasks, inspired by synaptic consolidation.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%91

Updated Jul 13, 2026 · TRV-2026-0210

68
ProblemPolicy· Stable· Evidence: Moderate (1 source)

Algorithmic systems used in employment screening and welfare administration reproduce historical disadvantage and generate new exclusion while eroding relational recognition and producing trust deficits.

Published 12 February 2026 in Societies, this peer-reviewed article analyzes how algorithmic systems in employment screening, welfare administration, and digital platforms function as social and institutional actors. Using regulatory materials, platform governance documents, technical disclosures, and composite vignettes synthesized from public evidence, it examines how automated classification and delegated authority reshape how individuals are evaluated and legitimised.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%91

Updated Jul 13, 2026 · TRV-2026-0202

68
ProblemPolicy· Stable· Evidence: Moderate (1 source)

The same survey found overreliance on labels, making participants more susceptible to false claims with human-made images and more hesitant to believe true claims illustrated with labeled AI-generated images.

Researchers studied whether legally mandated disclosure labels help users avoid deception from AI-generated images. After five focus groups, they surveyed 1,354 participants on how labels changed their judgments of true and false claims illustrated with different image types.

Impact 30%49
Evidence 25%95
Scale 20%35
Confidence 15%87
Recency 10%91

Updated Jul 13, 2026 · TRV-2026-0198

Recomputed live from the record · Aug 28, 2026, 6:28 AM