Straiker, a cybersecurity startup founded in 2025 and based in California, announced a $64 million Series A funding round that brings its total capital raised to $85 million. The round was co-led by Marathon Management Partners, Citi Ventures, Illuminate Financial, and Workday Ventures, with participation from Bain Capital Ventures and Lightspeed.

The company’s platform targets a specific and growing problem: organizations deploying AI agents, whether internally built or sourced from third parties, frequently lack visibility into what those agents can access, how they behave, and what risks they introduce. Straiker’s offering combines several capabilities designed to address this across the full agent lifecycle.

How the Platform Works

  • AI discovery: Identifies AI agents operating within an organization’s environment and maps their access and behavior.
  • Pre-deployment adversarial testing: Surfaces vulnerabilities before agents reach production, informed by threat detections observed at runtime.
  • Runtime protection: Stops threats in real time during production, with protections continuously refined by findings from pre-deployment testing.

The feedback loop between testing and runtime defense is central to the platform’s design. Findings from production deployments sharpen pre-deployment tests, while flaws uncovered in testing harden runtime controls. Straiker also works directly with frontier AI labs to gain early intelligence on emerging attack techniques, which the company uses to update its detections and defenses.

Co-founder and CTO Sreenath Kurupati described the platform’s foundation as pairing a comprehensive agentic exploit dataset with a security engine built specifically for autonomous threats, positioning it as purpose-built for enterprises that rely on AI agents at scale.

The funding arrives as enterprise adoption of AI agents accelerates and security teams face a gap between deployment speed and available tooling. Straiker’s approach reflects broader industry recognition that traditional security controls were not designed with autonomous, decision-making software agents in mind.