A 16-year-old security researcher known as Faav has disclosed a critical authentication flaw in Microsoft’s internal Titan analytics service that could have let an attacker run administrator-level SQL queries against an environment estimated to contain 17.3 trillion stored rows.
The issue was not a confirmed customer-data breach. Faav said the potential impact was hypothetical and that he limited testing to metadata, table descriptions, and tightly bounded one-row samples.
He reported the issue to the Microsoft Security Response Center (MSRC), which locked down the affected API endpoint within days.
16-Year-Old Researcher Finds Microsoft Flaw
Titan’s web interface was protected behind a “VPN REQUIRED” page, but its backend API was publicly reachable through a separate Azure-hosted endpoint.
A publicly exposed Swagger document listed several API routes, including /v2/Query, an endpoint capable of processing raw SQL queries.
Although the API expected Azure AD bearer tokens, Titan reportedly validated fields in a JSON Web Token, including the tenant ID, audience, application ID, and user claim, without validating the cryptographic signature that proves a token was genuinely issued by a trusted identity provider.
This flaw enabled the researcher to create a synthetic token using the alg: none setting, which indicates an unsigned JWT. After modifying the upn claim to admin, Titan allegedly resolved the value to a local administrator account and executed the submitted SQL query.
The flaw highlights a dangerous authentication anti-pattern: validating untrusted token claims before verifying the token’s signature. A token’s tenant, audience, or username fields cannot be trusted unless the application first confirms its cryptographic integrity.
Faav tested 56 historic routing values recovered from archived Titan configuration files and found that 30 remained active. Those values mapped through 24 configurations to 17 connected ClickHouse analytics databases containing 9,863 unique table names.
Using database metadata, including system.tables.total_rows and active system.parts, the researcher estimated the accessible environment contained 17,333,335,124,315 rows.
However, the number should not be interpreted as 17.3 trillion unique users, customers, or leaked records. The estimate likely included historical records, duplicated data, replicas, and derived analytics datasets. The researcher emphasized that he did not bulk-extract the data or access customer PII.
The metadata reportedly exposed approximately 25,000 account and email records, 17,990 employee email records, 15,001 employee organization records, thousands of dashboard and dataset definitions, and configuration information for connected databases.
Faav also retrieved two bounded one-row samples from a Bing analytics partition to validate that such data was technically reachable. The researcher said his self-developed AI hacking assistant, named Antares, initially discovered the exposed Titan API on August 25.
The system enumerated subdomains, examined the API schema, tested authentication behavior, and progressively identified how Titan processed altered JWT claims.
The automated process stalled because it treated the upn claim as a conventional email-formatted Entra identity. Faav’s manual intervention led to the pivotal test: changing the claim value to admin. That result demonstrated how human intuition and automated reconnaissance can combine in vulnerability research
Microsoft requested that testing stop while it investigated, then locked down the API endpoint on September 9. The company awarded the researcher a $5,000 bounty on September 17 and said the coordinated disclosure helped it harden its services.
For developers, the lesson is straightforward: applications must cryptographically verify JWT signatures before trusting any claim inside the token. Tenant validation, audience checks, and role lookups do not provide security when an attacker can forge the token that carries those values.
Join 16,000+ SOC teams using ANY.RUN to streamline threat investigations and reduce manual effort. Explore for your team
