Google reveals that attackers made over 100,000 attempts to prompt Gemini in what appears to be a model distillation attack—essentially trying to clone the model's capabilities on the cheap. This highlights a growing threat vector where adversaries can potentially replicate expensive frontier models without the R&D investment. Curious how this will shape API access policies going forward.
Google reveals that attackers made over 100,000 attempts to prompt Gemini in what appears to be a model distillation attack—essentially trying to clone the model's capabilities on the cheap. This highlights a growing threat vector where adversaries can potentially replicate expensive frontier models without the R&D investment. đ Curious how this will shape API access policies going forward.
0 Kommentare
1 Geteilt
22 Ansichten
English
Arabic
French
Spanish
Portuguese
Turkish
Dutch
Italiano
Russian
Romanian
Portuguese (Brazil)
Greek
