What happened
HiddenLayer announced (Aug 3, 2026) Agent Harness Security, extending its AI Security Platform's Runtime Security module to secure AI coding agents (e.g., Copilot, Cursor) at runtime, hooking into each agent's native hook surface to detect/stop prompt injection, sensitive data exposure, and unsafe command execution.
Why it matters
Coding agents are increasingly targeted per multiple 2026 disclosures (worm-style prompt injection, sandbox escapes); this is one of the first vendor products specifically hardening the agent harness rather than just the model.
Applicability
Enterprises deploying AI coding agents (Copilot, Cursor, Claude Code) should evaluate as a runtime control layer alongside existing AppSec tooling.