Teen Researcher Finds 17 Trillion Microsoft Records Exposed via JWT Flaw

Overview

A 16-year-old independent security researcher has disclosed a critical vulnerability in Microsoft’s internal infrastructure that potentially exposed 17.3 trillion rows of data. The flaw, discovered in an internal analytics service known as Titan, allowed attackers to forge authentication tokens and execute unauthorized SQL queries.

The researcher, who goes by the handle Faav, identified the issue in August 2026 after ten days of automated probing. While the vulnerability granted the ability to impersonate administrators and query sensitive datasets, Faav confirmed that no customer data or personally identifiable information was accessed during the investigation.

“Their submission and coordinated vulnerability disclosure helped us to better protect our customers by hardening our services.” — Microsoft Security Response Center

The Discovery

Faav’s journey began with Antares, a personal AI-driven hacking assistant developed to automate vulnerability detection. On August 25, 2026, Antares identified an internal Microsoft service called Titan. Although the web interface was protected by a mandatory VPN gateway for employees, the underlying API endpoints were accessible from external networks.

Initial reconnaissance revealed a Swagger documentation file listing four primary routes. Among them, the /v2/Query endpoint stood out because it accepted raw SQL inputs. Unlike the other routes, which utilized Azure AD bearer authentication, this specific endpoint appeared to handle queries for tabular data.

The “VPN REQUIRED” page shown to a non-employee visiting Titan’s frontend.

To understand the scope of the service, Faav utilized the Wayback Machine to recover archived configuration snapshots from 2023. This provided 56 table definitions, including a routing value labeled TestData. Armed with this metadata, Faav began testing the query endpoint using synthetic requests.

Breaking the JWT Authentication

The first barrier was authentication. Requests without a valid token were rejected with a 401 Unauthorized error. Over the following ten days, Antares systematically probed the JWT (JSON Web Token) validation logic. The bot cycled through various error states, adjusting tenant IDs, audiences, and application IDs until it reached a user lookup stage.

The critical breakthrough came when Faav noticed that the service accepted tokens with changing claims while the signature remained constant. This behavior indicated that the backend was validating the payload contents but failing to verify the cryptographic signature.

Faav exploited this by generating a synthetic JWT with the algorithm set to none. The header looked like this:

{
  "alg": "none",
  "typ": "JWT"
}

Because the signature field was left empty, the token consisted of only two base64url-encoded sections separated by a period. Titan accepted this malformed token as valid, effectively bypassing all identity verification checks.

Escalating to Administrator Access

With an unsigned JWT, Faav could inject arbitrary claims into the token. The next step was determining which user identity the backend would recognize. Antares tested numerous placeholder values and service aliases, but the system consistently returned “User not found.”

The final piece of the puzzle came on September 5, 2026, during a late-night session. Realizing that the backend might be using the upn (User Principal Name) claim as a local username rather than a standard email identity, Faav replaced the email-formatted UPN with the string admin.

This simple change resulted in a successful response. The system accepted the forged token and treated the requester as an administrator. From this point, Faav could submit SQL queries against the linked datasets.

I’d exploited an unsigned JWT bypass by hand before I ever used AI, so I recognized the pattern immediately. — Faav

Scope and Impact

Using only table descriptions, metadata, and bounded sample rows, Faav estimated the potential scope of the exposure. The Titan service aggregated data from a wide range of Microsoft internal datasets. The total volume of reachable rows was calculated at approximately 17.3 trillion.

This included sensitive analytics data related to Microsoft’s internal operations. However, Faav emphasized that the impact described is hypothetical. The researcher did not extract actual customer records or PII. Instead, the proof-of-concept relied on metadata analysis to demonstrate the severity of the flaw.

The vulnerability highlights a common but severe misconfiguration in internal enterprise tools: trusting JWT claims without verifying signatures. By setting the algorithm to none, attackers can often trick legacy or poorly configured parsers into accepting forged identities.

Disclosure and Response

Faav reported the vulnerability through Microsoft’s Bug Bounty Program. The company acknowledged the finding and credited the researcher for the coordinated disclosure. Microsoft stated that they have since hardened their services to prevent similar attacks.

We appreciate the opportunity to investigate the findings reported by Faav. Their submission and coordinated vulnerability disclosure helped us to better protect our customers by hardening our services.

This discovery marks Faav’s second major finding against Microsoft. The researcher previously published a report in 2025 detailing how guest check-in systems could be exploited to leak PII. At just 16 years old, Faav has established a reputation for combining traditional bug bounty hunting with AI-assisted automation.

Microsoft confirmed that they retain editorial control over technical write-ups to ensure no active exploitation vectors are published prematurely. While some figures and sections were modified prior to publication, the core findings regarding the JWT flaw remain unchanged.

Conclusion

The case of the 17 trillion exposed records serves as a stark reminder of the risks associated with internal API security. Even services hidden behind VPNs can be vulnerable if authentication protocols are not rigorously enforced.

For developers and security teams, the incident underscores the importance of:

Validating JWT signatures regardless of the algorithm header.
Implementing strict allowlists for internal service endpoints.
Monitoring access logs for unusual authentication patterns.

As AI-assisted security research becomes more prevalent, vulnerabilities like this may be discovered more frequently. However, robust security practices and responsible disclosure remain the best defenses against large-scale data exposure.