Defender Impact
Security teams have long struggled with alert volume that obscures compound risk — individually tolerable issues that combine into critical exposures. Northstar Security’s AI-agent platform targets exactly this gap by correlating identity, asset, vulnerability, and business-system data across the enterprise into a live graph that surfaces toxic combinations before they are exploited.
Capability Overview
Northstar Security emerges from stealth with $7 million in seed funding and a platform built around what the company calls a Security Context Graph. The graph ingests telemetry from more than 70 enterprise security and IT products — including Okta, Microsoft Defender, Entra ID, GitHub, Workday, AWS, and Google Cloud — and maintains a continuously updated map of how identities, assets, applications, vulnerabilities, and business systems relate to one another.
The core defensive proposition is toxic-combination detection. A single vulnerability on a laptop, an unmanaged device, or an account with excessive privileges may each sit below the remediation threshold in isolation. Northstar’s graph links these signals: a flagged example from the company’s own materials describes a laptop lacking endpoint protection whose owner holds privileged access to a payroll system — a path that siloed tools would never surface as a unified finding.
Analysts interact with findings through a natural-language interface, allowing plain-English queries against the graph without requiring deep technical familiarity with underlying data schemas. AI agents can then initiate remediation — installing endpoint protection, restricting access, or notifying the affected user — with human supervision available at configurable points in the workflow. The platform also exposes the graph to external AI agents via Model Context Protocol (MCP), enabling organisations to compose bespoke security workflows on top of Northstar’s data layer.
The founding team brings relevant pedigree: CEO Amer Deeba previously co-founded Normalyze (acquired by Proofpoint) and held senior roles at Qualys; CTO Ed Rowe led engineering at AppDynamics and Farmers Business Network.
Defensive Advances
Cross-silo risk correlation at scale. Defenders can now identify multi-factor exposure paths that span identity, endpoint, cloud, and application data in a single view — something that previously required bespoke SIEM logic or manual analyst correlation across dashboards.
AI-accelerated remediation. Agents can execute targeted fixes (access restriction, agent deployment, user notification) without requiring analysts to context-switch across multiple consoles, compressing mean-time-to-remediate for compound findings.
Accessible investigation for non-specialist analysts. Natural-language querying of a complex graph lowers the skill bar for tier-1 and tier-2 analysts to investigate sophisticated, multi-system risk paths — partially addressing the talent gap that makes deep graph analysis impractical at most organisations.
Composable workflow extension via MCP. Exposing the Security Context Graph to external AI agents means security engineers can build custom detection and response logic on top of Northstar’s correlated data without waiting for vendor-native feature development.
Residual Gaps
Several maturity questions shape how quickly organisations can realise full value from this platform:
- Graph completeness depends on integration breadth. The 70+ connector catalogue is strong for a seed-stage product, but organisations operating niche or legacy tooling not yet supported will have incomplete graph coverage — meaning toxic combinations involving those systems will remain invisible.
- AI-driven remediation governance is not trivial. Enabling agents to restrict access or modify configurations at scale requires mature change-management and rollback processes. Organisations without those controls in place should treat autonomous remediation as a future-state capability and supervise all actions initially.
- Overreliance risk on graph accuracy. If source tool integrations are misconfigured or data is stale, the graph’s confident-looking findings may reflect an outdated reality. Establishing data-quality validation and monitoring for the graph itself is a prerequisite for trusting agent-initiated actions.
- MCP governance is an emerging discipline. Exposing the Security Context Graph to external agents via MCP introduces agentic workflow complexity that most security teams have not yet operationalised safely.
Framework Mapping
Northstar’s capability is most directly relevant to AML.T0012 (Valid Accounts) — its graph explicitly surfaces over-privileged accounts as components of toxic combinations. The MCP-based agent access layer warrants attention to AML.T0086 (Exfiltration via AI Agent Tool Invocation) and AML.T0110 (AI Agent Tool Poisoning) as the organisation scales its agentic workflow surface. On the OWASP side, LLM08 (Excessive Agency) and LLM09 (Overreliance) are the primary maturity considerations when configuring autonomous remediation thresholds.
Deployment Considerations
Organisations should approach Northstar in phases: begin with read-only graph ingestion and natural-language investigation to validate finding quality and integration coverage before enabling any agentic remediation. Prioritise connecting identity providers (Okta, Entra) and endpoint/vulnerability management tools first — these generate the highest-value toxic-combination signals. Establish explicit human-approval gates for remediation actions involving access restriction or configuration change, and treat MCP-based external agent access as a controlled experiment requiring dedicated governance before broad enablement.
Defender Checklist
- Map your current tool estate against the 70+ Northstar connector catalogue and identify coverage gaps
- Define risk-combination scenarios most relevant to your critical assets and validate that Northstar surfaces them accurately in a pilot
- Set human-approval thresholds for every remediation action type before enabling autonomous agent execution
- Establish a data-quality monitoring process for the Security Context Graph itself
- Engage your change-management team before enabling access-restriction remediation at scale
- Review MCP governance requirements before exposing the graph to external AI agents