If you’re a big AI shop pushing open-weight models but not sharing the training data then you can easily introduce targeted blind spots.

What happens when nation state-backed labs push out state of the art open weight models so that people embed them in their applications with intentional weaknesses?

Reminder: if the service (or model) is free then you’re the “product”. Or in this case you’re the target.

#ai #LLM #cybersecurity