Trust infrastructure requires transparency. Here's how we handle reports, moderation, and provider responses.
1458
Total reports
100%
Publish rate
649
Verified this week
0%
Provider response rate
Submit
Users submit incident reports with evidence, AI output screenshots, and context.
Review
Moderators verify claims, check for PII, and assess severity within 72 hours.
Publish
Verified incidents are published to the public record with full anonymisation.
Provider response
AI providers can submit an official response visible alongside the incident.
ALPAR AI calculates a Trust Score for each AI provider on a 0-100 scale to quantify their transparency and accountability to the public.
Negative impact. Each verified incident reduces the provider score.
Positive impact. Prompt official responses to incidents increase score.
Community evaluations of model features and performance.
Credits given for publishing public bias, privacy, and safety audits.
ALPAR AI holds the most comprehensive independent dataset of real-world AI failures, accessible to researchers, regulators, and the public.
K-BENCHMARK is ALPAR AI's open methodology for scoring AI models across key safety dimensions. We test models based on real-world incidents, not just academic tests.
ALPAR AI is committed to operating with full transparency. Every moderation decision follows a documented process, and aggregate statistics are published monthly.
We do not accept payment to alter incident records. Providers may submit factual corrections through the official response process only.
Our moderation team operates independently. No AI provider has the ability to suppress or alter published incident reports.
To guarantee trust and institutional readiness, ALPAR AI is pursuing industry-standard compliance certifications.
Target: Q3 2026. Setting up continuous security logging, audit trails, and strict data access policies.
Target: Q1 2027. Implementation of an Information Security Management System (ISMS) across all pipelines.