Northstar Security leaves stealth with $7m and AI agents

A security startup called Northstar Security came out of stealth on 1 November 2026 with $7 million in funding, according to SecurityWeek. The company says its product maps how identities, assets and vulnerabilities connect to one another, and uses AI agents to find and fix risky combinations of issues.

SecurityWeek describes the target problem as "toxic risk combinations". In security work, that phrase generally refers to a set of individually unremarkable weaknesses that become serious when they line up: an over-permissioned account, for example, attached to a machine that is reachable from the internet and running software with a known flaw. Scanners that grade each issue separately tend to rate all three as middling and leave defenders to join the dots themselves.

The detail that matters for readers of this paper is the delivery mechanism. Northstar is not only using AI to score or rank findings. SecurityWeek reports that the company uses AI agents both to find the risky combinations and to fix them. Remediation is the step where autonomy stops being a reporting feature and starts changing production systems. An agent that can revoke a credential, tighten a policy or patch a host is an agent that holds privileges worth attacking, and whose mistakes are felt immediately rather than filed in a backlog.

The published material does not say how much autonomy the agents are given, whether fixes require human approval, which cloud or identity platforms are supported, or how the agents themselves are constrained and audited. Nor does it name customers, say whether the product is generally available, or identify the investors behind the $7 million. Those gaps are worth noting because the security value of an autonomous remediation tool depends almost entirely on the answers.

Northstar arrives in a crowded space. Exposure management, cloud security posture tools and identity governance products all claim some version of correlating issues across identities and assets, and a growing number now advertise agentic features. The differentiator claimed here, on the evidence of the announcement, is the combination of a connected graph of identities, assets and vulnerabilities with agents that act on it.

For security teams evaluating this class of product, the practical questions are familiar. What permissions does the agent hold in your estate, and can they be scoped down? Is every action logged in a form your auditors will accept? Can a change be rolled back quickly if the agent misreads a dependency? And what happens when the data the agent reasons over, such as ticket text or asset descriptions, contains instructions planted by someone else?

The Agentic Times has not independently tested the product. Further detail on the architecture, the guardrails and the funding round will determine how much weight the "find and fix" claim can carry.