regnil
Lukasz OlejnikInserting backdoors into AI/LLM models. Challenge: how to determine if the model implemented in a company, a government institution or the army does not have a backdoor and is safe? Not known for 100%. The beauty of artificial intelligence.
"we believe that our results demonstrate two key conclusions: both of
our threat models are possible and they could be very difficult to deal with if they did occur"
#ai #cybersecurity #artificialintelligence
https://arxiv.org/pdf/2401.05566.pdf
"we believe that our results demonstrate two key conclusions: both of
our threat models are possible and they could be very difficult to deal with if they did occur"
#ai #cybersecurity #artificialintelligence
https://arxiv.org/pdf/2401.05566.pdf
