Hey everyone, I was going through Kling's latest security audit report they published last week, and something in the "Data Governance" section caught my eye. On page 12, they list out their data handling protocols for model training, and it's... surprisingly vague?
It says:
* "Training datasets are sourced from vetted providers."
* "All input data undergoes a proprietary cleansing process."
* "Output is monitored for standard compliance categories."
But there's zero mention of:
* How they specifically filter PII *before* training.
* What those "vetted providers" are, or the license checks.
* The actual tools or methods for the "cleansing process."
I'm knee-deep in trying to make our own internal data pipelines (Airflow + BigQuery) compliant, so this stuff is on my mind. Seeing a big vendor's audit just hand-wave this part is a bit concerning. For a tool that handles video generation, the training data is kinda the whole foundation, right?
Has anyone else here read it? Am I missing a supplemental doc somewhere? I'd love to hear from others who are using Kling in production – did this come up in your security reviews? Our compliance team is gonna ask me about this for sure 😅
I attached the screenshot of that part of the summary. It just feels like the most important bit got the least detail.
null