Teen's AI hackbot finds Microsoft auth flaw in internal analytics

A 16-year-old security researcher using an AI "hackbot" he built himself gained administrator access to Microsoft's internal Titan analytics service, reaching databases holding an estimated 17.3 trillion rows, The Register reported on 30 September 2026. Microsoft has since locked down the affected API and paid the researcher, who goes by Faav, a $5,000 bug bounty.

Titan is an internal analytics platform whose web interface Microsoft restricts to its own employees, according to The Register. The researcher reached its API through an Azure Cloud Services host because Titan did not check the signature on a login token.

The work began on 25 August 2026, when the researcher's bot, named Antares, found Titan's public API. Over the following ten days the pair probed the service's JSON Web Token checks \u2014 JWTs are the signed tokens many services use to prove who a caller is \u2014 and its email-formatted user principal names. They found an unsigned token that could reach Titan's local user lookup, but no user name that Titan recognised.

Early on 5 September 2026 the researcher changed the unsigned token's user principal name from an email-formatted identity to "admin". Titan treated it as a local username, resolved it to local user ID 1, which carried an admin role, and allowed SQL queries to run. The core failing, according to the researcher, was that Titan validated the token's contents \u2014 tenant, audience, app ID and user \u2014 but never verified the signature. He compared it to a hotel where every door has a working keycard reader but any keycard opens any room. His stated takeaway for developers and coding agents alike is to verify signatures above all else when building authentication.

The access opened Titan's platform metadata database and, from there, application tables. The metadata included around 25,000 account and email records, 17,990 employee email records, 15,001 employee organisation records, 355 database configurations, 20,979 virtual-dataset SQL definitions, 24,569 dashboards, 425,891 charts and 27,347 dataset definitions. The user and usage directory exposed employee job titles, departments and management hierarchy, which the researcher said could be useful for social engineering, adding: "though I never tested or demonstrated that". He also found a Bing analytics sample and tested two rows containing search information, identifiers and country- or state-level location details, saying the location values were not precise.

Testing 56 routing values from an archived configuration, he found 30 still active. Those resolved through 24 configurations to 17 connected analytics databases spanning 9,863 unique table names, producing the 17.3 trillion figure. He described it as a storage estimate derived from metadata that likely includes historical, duplicated and derived data.

Between 6 and 8 September 2026 Microsoft asked the researcher to stop testing and requested his IP address to confirm there had been no activity beyond the bounty research. The company locked down the endpoint a day later and told him the "report prompted immediate investigation and remediation to address the remaining exposure". The bounty was awarded on 17 September 2026.

"We appreciate the opportunity to investigate the findings reported by Faav," Microsoft said in a statement provided for his blog, adding that the coordinated disclosure "helped us to better protect our customers by hardening our services". The researcher said he rewrote his blog post at Microsoft's request, cutting sections and numbers and rewording the impact before publication.