Summary

  • HiddenLayer announced a $100 million Series B led by Delta-v Capital on September 2, with agent-runtime protection among its investment priorities.
  • Its coding-agent product explicitly ties enforcement to the capabilities of the host platform. Detecting an action and having authority to stop it are different deliverables.

For an enterprise buying protection for AI coding agents, a useful demonstration ends with a stopped action, not merely a detected threat. HiddenLayer’s new financing puts more capital behind that distinction. The company’s September 2 announcement says its $100 million Series B will support expansion of its enterprise platform, including runtime security and Agent Harness Security.

The latter is not a September launch. HiddenLayer introduced it on August 3, describing integration with the native hooks through which coding agents expose points of intervention. The commercial proposition is to place security within software work as it happens, where an agent may handle source files, tool responses or commands.

That position matters because an alert and an intervention arrive at different moments. A record of an unsafe operation can help an investigation. Preventing the operation requires a control point before it takes effect. Redacting sensitive context before it reaches a model is a different action again. A buyer comparing products needs to know which of these outcomes is available for the work being purchased.

HiddenLayer makes the dependency explicit. Its product description qualifies inline blocking and redaction by the underlying platform’s capabilities, and promises visibility into configured versus enforced policies. This is an important boundary on the offering, not evidence that the product is defective. A common policy can encounter different execution rights on different hosts.

GitHub’s own Copilot hooks reference shows how consequential such details can be. A command-based pre-tool hook denies the tool call on a non-timeout error, but a timeout returns it to the normal permission process. HTTP pre-tool hooks also return to that process on network errors, timeouts or unsuccessful HTTP responses. Returning to ordinary permissions does not mean unconditional permission to run.

Those are GitHub’s documented mechanics, not a finding about which integration HiddenLayer uses. The reviewed public material does not establish the vendor’s exact hook variant for each agent, or provide an independent benchmark of delay and false positives. The comparison identifies questions for a deployment review; it does not establish a bypass or a security incident.

The funding therefore strengthens an expansion plan without settling product acceptance. For a customer, the relevant unit of coverage is an action on a particular agent platform under a known policy and failure condition. Counting supported product names alone leaves that unit unspecified.