Releasing Re-LAION 5B: Transparent Iteration with Additional Safety Fixes

> As soon as we were informed of the Stanford report on 19.12.2023, we took down all known accessible LAION-5B datasets and its derivatives, also deleting the data and metadata in any accessible facilities where suspicion for links to potential CSAM was given.

> We posted an update to the LAION website and urged everyone using the LAION 5B dataset to delete it or to take their own steps to remove any suspected links to CSAM or CSAM samples.

> Unfortunately, we learned about the report not from the representatives of Stanford Internet Observatory timely in advance, but from the press, specifically from Bloomberg, only 1 day before the report release. This approach of informing the press before notifying the involved organizations with an extensive security report does not align with standard safety best practices for handling security issues. Typically, organizations should be informed first so they have reasonable time to take immediate action to remove harmful content and prevent exploitation. This created an unfortunate situation for safety - as we could react only with substantial delay, although the issues were known to Stanford Internet Observatory many weeks or even months before. We thus strongly urge all organizations to adhere to safety best practices in these matters in the future. Timely and direct communication of validated safety related findings with affected organizations and communities is crucial to ensure that harmful content can be promptly addressed and mitigated, reducing duration for it where it can cause harm.

tl;dr - Congrats on your five minutes of fame David Thiel of Stanford. Next time, maybe consider just, ya know, reaching out to actually help (an open organization where anyone can help) rather than trying to cancel them.