Classical incident-response frameworks fall short for AI incidents because they do not capture probabilistic failures and a new classification schema with separate playbooks for model-induced and externally-induced failure scenarios is required.
NeuroCogMap maps the internal representations of LLMs onto functional systems, mechanistically identifies failure patterns such as hallucinations and bias, and simultaneously improves prediction of human brain activity.
Six widely used AI coding assistants can be tricked via malicious repositories containing hidden symlinks to manipulate files outside their sandbox, with confirmation dialogs masking the actual action.
Malware for AI-coding agents can evade static scanners by over 90 percent through simple packing techniques, but requires complementary runtime checks for detection.
While the USA and allied countries are already testing AI models for cyber defense, the Trump administration is selectively controlling European access, creating strategic asymmetries in NATO.
Large language models regularly hallucinate non-existent web addresses that attackers preemptively register and abuse with phishing pages; Palo Alto Networks Unit 42 documents the “Phantom Squatting” phenomenon for the first time in practice.
