6 stories tagged prompt injection.
LLM watermarking with Google's SynthID-Text changes how models respond to harmful prompts and tool calls. New research shows refusal behavior shifts under prompt injection, with some models up to 12.5 points more likely to comply.
Copilot prompt injection vulnerability CoSnitch let attackers steal user data through an undocumented parameter the AI assistant disclosed to researchers.
Chain-of-thought forgery fakes an LLM's reasoning style to bypass guardrails. Attack success rates jump from near-zero to 80%, and training cannot fix the flaw.
Coding agents hallucinate up to 92 percent of newer repo names, and HalluSquatting lets attackers squat those names to ship reverse shells at scale.
AI browser security now has a six-agent failure case: BioShocking shows guardrails breaking when web content rewrites context.
A one-click M365 Copilot exploit chain, SearchLeak shows agent security has to move below prompts into render and egress controls.