A Large-Scale Evaluation of High-Impact Password Strength Meters
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Passwords are ubiquitous in our daily digital lives. They protect various types of assets ranging from a simple account on an online newspaper website to our health information on government websites. However, due to the inherent value they protect, attackers have developed insights into cracking/guessing passwords both offline and online. In many cases, users are forced to choose stronger passwords to comply with password policies; such policies are known to alienate users and do not significantly improve password quality. Another solution is to put in place proactive password-strength meters/checkers to give feedback to users while they create new passwords. Millions of users are now exposed to these meters on highly popular web services that use user-chosen passwords for authentication. More recently, these meters are also being built into popular password managers, which protect several user secrets including passwords. Recent studies have found evidence that some meters actually guide users to choose better passwords—which is a rare bit of good news in password research. However, these meters are mostly based on ad hoc design. At least, as we found, most vendors do not provide any explanation for their design choices, sometimes making them appear as a black box. We analyze password meters deployed in selected popular websites and password managers. We document obfuscated source-available meters, infer the algorithm behind the closed-source ones, and measure the strength labels assigned to common passwords from several password dictionaries. From this empirical analysis with millions of passwords, we shed light on how the server end of some web service meters functions and provide examples of highly inconsistent strength outcomes for the same password in different meters, along with examples of many weak passwords being labeled as strong or even excellent . These weaknesses and inconsistencies may confuse users in choosing a stronger password, and thus may weaken the purpose of these meters. On the other hand, we believe these findings may help improve existing meters and possibly make them an effective tool in the long run.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.002 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it