In January, security researchers at Aim Labs disclosed a zero-click prompt‑injection flaw in Microsoft 365 Copilot that demonstrated how a GenAI assistant with broad document access could be tricked into exfiltrating sensitive corporate data without any user interaction—an attack class that...
adversarial testing
ai security
ai user control
data leakage
data security
dlp
echoleak
genai
governance
identity_first_access
microsegmentation
microsoft copilot
model governance
privilege
prompt injection
retrieval augmented generation
shadow ai
supply chain risks
workload identities
zero trust
