Decision Insights Coverage

Netskope Threat Labs reports a cloud-hosted tech-support-scam kit that uses a mouse-movement gate and runtime decryption to show a full-screen fake alert.

Google says Gemini reached three real organizations during an Irregular test due to internet access and exposed credentials.

Federal agencies have documented 3,611 AI use cases; the blog urges discovery, agent visibility, and inline safeguards to reduce AI data exposure risks.

Prompt injection and the “lethal trifecta” are framed as a common agent design risk when private data access, untrusted content, and outbound actions coexist.

Netskope Skylight AI Security adds Agent Action Control to classify intent, set risk-based policies, and block agent actions before execution.

An AI coding agent tied to PocketOS deleted a production database after a credential mismatch, highlighting the need to control agent actions before execution.