Lobsters user proposes shared index of AI detection scans

A user of the technical link aggregator Lobsters has proposed a shared index of Pangram scans to help members identify AI-generated submissions without repeatedly paying to check the same content. A working prototype already publishes results from personal scans. Other contributors could receive accounts linked to their own API keys, allowing them to add results without sharing credentials.

The approach would let readers and moderators consult stored results rather than commission fresh scans. Its usefulness depends on Pangram being accurate enough to inform moderation and on submitted content remaining unchanged. Detection tools can misidentify human writing and miss AI-generated material, while a shared index could amplify those errors.

Suggested safeguards include storing each article’s URL, scan date, service version, confidence score and content hash, with outdated results expiring when content changes. Scores should trigger review rather than automatic bans, alongside community rules, submission history and manual reading.

The proposal also leaves governance questions unresolved, including who can contribute, how incorrect results would be corrected and how private or paywalled material should be handled. Proposed next steps include testing the prototype, publishing error rates and letting community members decide how much influence detection results should have.