Open, Closed and Broken: Prompt Fuzzing Finds LLMs Still Fragile Across Open and Closed Models

Palo Alto Unit 42 published a batch of active-threat and research reports covering two main themes: GenAI/LLM security and active exploitation of multiple high-severity vulnerabilities and nation-state activity. Research shows LLM guardrails and “AI judges” are fragile to scalable prompt-fuzzing and indirect/web-based prompt injection, enabling stealthy evasion of AI security controls. On the exploitation side Unit 42 documents active abuse of multiple critical flaws (Chrome Gemini panel, BeyondTrust RCE, Ivanti EPMM zero-days) with observed deployment of web shells, backdoors, VShell andSpark