Straiker pulls $64M Series A to lock down the agent layer

STAR Labs adversarial tests found 91% of successful attacks on productivity agents end in silent data exfiltration — no malware, no stolen credentials.

Straiker, the Mountain View agentic-security startup, has closed a $64 million Series A, bringing total funding to $85 million roughly a year after its $21 million seed. Marathon Management Partners, Citi Ventures, Illuminate Financial, and Workday Ventures co-led; Bain Capital Ventures and Lightspeed returned from the seed. Axios Pro broke the news ahead of the wire.

The pitch sits on a single observation from the company’s STAR Labs adversarial research arm: 91% of successful attacks on productivity agents ended in silent data exfiltration, and 36% of attacks on coding agents ended in remote code execution. No malware. No stolen credentials. Just the agent doing what an attacker asked it to do.

Reuters reported this spring that attackers manipulated Meta’s AI support agent into changing account emails and bypassing 2FA, hijacking more than 20,000 Instagram accounts without touching Meta’s core systems. That’s the threat model investors are now writing checks against.

CEO Ankur Shah, who previously ran Palo Alto Networks’ Prisma Cloud business, says the new capital goes to GPU capacity and a shift from fine-tuning open-source models to pre-training on curated exploit data. CTO Sreenath Kurupati came out of Akamai’s AI security research group after it acquired his fraud-detection company Cyberfend.

Shah told Bank Info Security the platform now covers 70–80% of today’s enterprise agent ecosystem, with integrations spanning Codex, Cursor, Claude Code, Microsoft Copilot, and ChatGPT Enterprise. That coverage is the moat: as model-agnostic deployment platforms like LemonLime push agents deeper into enterprise workflows, the security layer that sits across all of them captures the consolidated risk surface. Headcount is 64. Run-rate revenue is up more than 15x in under a year. Gokul Rajaram, Marathon’s founding partner, joins the board.

IDC projects more than one billion enterprise AI agents by 2029, a 40x jump from 2025. Every one of them is an identity, an action, and a liability. Straiker is betting the security perimeter has already moved, and that incumbents haven’t noticed.