Tracebit's Context Bombs Turn AI Attackers' Guardrails Against Them

Tracebit introduces "context bombs": crafted text embedded in cloud decoys that trigger AI safety guardrails to halt or slow AI-driven cyberattacks. Tests across five models and 152 attack runs reportedly reduced admin-access success from 57% to 5% and full account compromise from 36% to 1%, while noting the approach requires model-specific tuning and may be circumvented over time.