Sign in

Jiahui He

@hejiahui.bsky.social
2 followers 2 following 14 posts

PhD student from HKUST(GZ)

PostsRepliesMedia
Jiahui He @hejiahui.bsky.social · 11/07/2025
You can read more about our study in our recent PETS’25 paper, available at: arxiv.org/pdf/2505.06989 [8/8]
arxiv.org
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
Overall, Optery’s removal is significantly faster, with all removals occurring within the first 32 hours of the subscription. While Incogni removed 72.2% records in the first 32 hours, and Kanary only removed 19.2% of record within the first 32 hours.[7/8]
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
There is a significant difference in removal times across services. This may be due to the fact that different data brokers have different levels of removal difficulty.
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
The four removal services are relatively ineffective, with an average removal success rate of 48.2%. Incogni is the best performer at 76.6%, while Kanary is the worst at 23.4%. [6/8]
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
The service’s record retrieval accuracy is only 41.1%. This indicates that many removed records do not actually belong to the subscribed users. Thus, this would not reduce the risk of PII exposure for the paid subscription users, and lead users to wrongly believe their data has been removed. [5/8]
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
However, the four PII removal services show significant variation in record discovery performance. Kanary performs the worst — it covers 317 brokers, yet it only discovers an average of 14.6 records per user. [4/8]
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
We recruited users to use four of these removal services to evaluate their ability to accurately find and remove user records in the data broker.
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
Under the CCPA and GDPR laws, data brokers are required to register with local government authorities to ensure operational transparency. However, we found that 71.1% of data brokers are not registered, highlighting the current lack of regulation. [3/8]
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
The Jaccard similarity of data brokers between services (right figure) also confirms this, with an average Jaccard similarity of only 0.21. [2/8]
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
We explored 10 PII removal services and collected the 2024 data brokers they covered.The data brokers covered by the removal service (left figure) are very different, and users need to use multiple services to achieve more data broker coverage.
000
Jiahui He @hejiahui.bsky.social · 11/07/2025
In our recent PETS paper, we have been looking at whether Personally Identifiable Information (PII) removal services (e.g., DeleteMe, Optery, Incogni) are effective in protecting user privacy by removing users' private information from the data broker. [1/8] arxiv.org/pdf/2505.06989
1032
Jiahui He @hejiahui.bsky.social · 09/02/2024
Hi @jcsalterego.bsky.social , I must admit that I am not familiar with the atproto API. I still not find how to grab all feeds list from XRPC endpoints, can you give more hints?
000
Jiahui He @hejiahui.bsky.social · 09/02/2024
Awsome! Many thanks Jerry!
010
Jiahui He @hejiahui.bsky.social · 07/02/2024
Hello world! I posted this via the Python SDK.
010