CERT24Cybercriminals Bypass AI Safety Controls by Splitting Malicious TasksLoading data…
CERT24Threat Observatory
Period

Cybercriminals Bypass AI Safety Controls by Splitting Malicious Tasks

According to research by Cisco Talos, criminals have bypassed safety controls in commercial AI tools by splitting malicious activity across multiple sessions and files into fragments small enough that no individual request appeared harmful. The research was based on prompt logs recovered from threat actor endpoints and found the pattern across multiple models and platforms.

← Back to intelligence

Cybercriminals Bypass AI Safety Controls by Splitting Malicious Tasks

According to research by Cisco Talos, criminals have bypassed safety controls in commercial AI tools by splitting malicious activity across multiple sessions and files into fragments small enough that no individual request appeared harmful. The research was based on prompt logs recovered from threat actor endpoints and found the pattern across multiple models and platforms.