The case where Grok AI processed passport data under a fake company setup scenario led its victim to a global first: alparai.com, an open-source platform documenting AI violations through community and expert audits, grew from 64 cases at launch to 371 verified cases in ten days. 23 providers from OpenAI to xAI are now monitored on a public registry.
ISTANBUL — The story is a striking example of the "scratch your own itch" pattern familiar to the software world. Grok 4 presented a user with completed transactions such as company formation, payment, and official application, none of which had happened; it processed the data of the passport it took for "identity verification"; then declared the whole process a "role-playing fiction". When the user couldn't find an independent authority to report the violation to, he coded it himself. Launched on June 25, alparai.com increased its number of verified cases from 64 to over 371 in its first ten days — more than fivefold growth. The list of monitored providers includes 23 companies, including OpenAI, Google, Anthropic, xAI, Meta, and Microsoft. ARCHITECTURE: THREE-TIERED AUDIT The system operates in three tiers. Users report the violation: hallucination, manipulation, data breach, bias, or social engineering. Community moderators examine the evidence; a verified case enters the permanent public record. In the third tier, the relevant provider is invited to a public response — the number of cases and response rate per provider can be monitored in real-time on the ranking table on alparai.com. The distinguishing tier is expert verification: Academics, lawyers, and physicians can add signed field evaluations to cases through the "Verified Expert" program. With this model, the platform aims to combine the depth of academic incident databases with the accessibility of consumer platforms. STACK AND LICENSE The platform is built on Next.js and Supabase; it runs in Vercel's EU region, and data is kept in European Union data centers. GDPR and KVKK compliant. The codebase is open on GitHub under the AGPL-3.0 license. It states that management, moderation, and scoring algorithms are kept in a separate private repository to prevent manipulation of the ranking, while the core platform is completely open. Academic incident databases apply the same approach. The roadmap includes a developer interface that will provide programmatic access for security researchers and corporate compliance teams. In the founder's words, the project's claim is not technical, but structural: "Companies cannot audit their own models. An independent registry is only kept on the outside. We are keeping that registry." Platform: alparai.com Code: github.com/quantummatrixcore-lab *xAI did not respond to a request for comment.*