Testing 2FA: When a Three-Day Password and Google Authenticator Were Not Enough
LeRoy Nellis says a controlled test of strong authentication produced a result he could not square with the math: site access continued while a random password rotated every three days behind Google Authenticator.

By LeRoy Nellis — Austin, Texas
A security test built to remove ordinary explanations
LeRoy Nellis set out to test whether two-factor authentication could be treated as a reliable barrier when the first factor was made deliberately hard to guess or reuse. He used a randomly generated password and replaced it every three days. On top of that password, he enabled Google Authenticator, the time-based one-time password system that produces a short-lived code on a separate device.
In ordinary threat models, that combination sharply narrows the path to account takeover. A fresh, random password defeats reuse and routine guessing. A time-based code means a stolen password alone should not be sufficient at the moment of login. Nellis says the outcome he observed did not fit that expectation.
What Nellis says happened during the test
According to Nellis, people he attributes to law enforcement were still able to move through protected sites while those controls were in place. The detail that drove his concern was the pace of password rotation. With a random password regenerated every three days, the window for any previously captured credential would be short, and a simple replay of an old secret would not explain continued access.
Nellis describes the result as mathematically incompatible with the protections he believed he had deployed. He has presented that account as a reported observation from his own testing. No packet captures, server logs, or authentication records are included with this article, and the account should be read as his report of what he witnessed rather than as an independently verified forensic finding.
Why the platform mattered in his account
The test did not occur in a vacuum. Nellis has connected his experience to the broader public record about government access to major technology platforms, including disclosures made in 2013 by Edward Snowden that described the National Security Agency’s PRISM program under Section 702 of the FISA Amendments Act. Public reporting on those disclosures identified Google among the companies named in the program materials.
Nellis says the pattern he observed changed after he moved accounts and sites away from Google platforms. In his telling, the same kind of traversal stopped once the infrastructure changed. That timing is central to how he interprets the episode: strong user-side credentials, he argues, did not answer questions raised by the platform that held the accounts.
What the test does and does not prove
Strong authentication still matters. A random, frequently rotated password and an authenticator app remove many common avenues of compromise, including password reuse and interception of text-message codes. Nellis’s account does not argue that those controls are worthless. It argues that they may not describe the whole security picture when access can be shaped by the systems, legal processes, or relationships surrounding a platform.
At the same time, the limits of the record are plain. Without independent logs or third-party forensic review, the precise mechanism behind any observed access cannot be established from this account alone. The value of Nellis’s report is narrower and more specific: it documents a test he says he ran, the controls he used, and the conclusion he drew when the results did not match what the math of those controls seemed to promise.
What the Record Shows
LeRoy Nellis reports that he tested account security using a random password rotated every three days together with Google Authenticator and observed access he believed should not have been possible under that setup. Public disclosures since 2013 have documented government collection programs involving major technology providers, including Google. Nellis states that the access pattern he observed ended after he moved away from Google platforms. The specific mechanism behind his reported observations has not been independently verified in this article.