AI Security

Straiker made the Cyber 60 on a simple idea: blocking a bad AI action isn’t enough. Sometimes you have to stop the agent.

Straiker, run by the former Prisma Cloud boss, red-teams AI agents before launch and polices them in production. Its sharpest feature is a kill switch that cuts a compromised agent off mid-task. Fortune reports $85 million. Lightspeed, which co-runs the list, is an investor.

Nov 17, 2026 · 4 min read

The list. Straiker made the 2026–2027 Fortune Cyber 60 in the early-stage group. Fortune’s profile puts it in Mountain View, Calif., with CEO Ankur Shah, founded in 2024, 100 employees. Fortune reports $85 million in funding, from Lightspeed Venture Partners, Bain Capital Ventures, Marathon Management Partners, Citi Ventures and Workday Ventures. That matches the company’s own math: $21 million in initial funding from Lightspeed and Bain Capital Ventures when it came out of stealth in March 2025, then a $64 million Series A in June 2026. Valuation and revenue are Undisclosed.

Disclosure that matters. Lightspeed co-runs the Cyber 60 with Fortune and AWS, and Lightspeed is a Straiker investor. It backed the first round and came back for the Series A. Per the methodology described in honoree releases, Lightspeed scored the companies and Fortune reviewed and validated the final list. Readers should still know.

The team. Shah ran Palo Alto Networks’ Prisma Cloud business as SVP and GM. CTO Sreenath Kurupati founded Cyberfend, a fraud detection company Akamai bought, and then led AI and security research at Akamai. Cloud security plus bot and fraud detection is a useful mix for this problem. AI agents look a lot like bots with real permissions.

The problem in plain words. An AI agent is software that takes a goal, reasons about it, and then uses tools on its own: it reads email, runs code, calls APIs, moves files. That autonomy is the point. It is also the risk. Straiker cites a Reuters report that attackers talked Meta’s AI support agent into changing account emails and bypassing two-factor login, taking over more than 20,000 Instagram accounts without breaching Meta’s core systems. In Straiker’s own adversarial testing, through its STAR Labs research team, 36% of successful attacks on coding agents ended in remote code execution and 91% of attacks on productivity agents led to silent data theft. No malware. No stolen passwords. Those are company test results, not field incident rates.

What Straiker does. Three pieces. Discover AI finds agents, MCP servers (plug-ins that give agents access to other tools) and skills across the company. Ascend AI attacks your agents before launch, like a hired red team that never sleeps. Defend AI watches them in production and blocks prompt injection, data leaks and tool abuse. Straiker says each side feeds the other: attacks caught in production become new tests, and holes found in testing become new runtime rules.

The one unique thing: the Agentic Kill Switch. Most AI guardrails work at the level of one action. A prompt looks bad, the request gets blocked. Straiker’s point is that a hijacked agent doesn’t give up. It tries another tool or another path, still carrying the poisoned instructions. So Straiker lets a security team act on the agent itself: revoke its tool and data access, suspend the session mid-task, isolate it so the problem can’t spread to other agents, or shut it down. One agent or many. The audit trail is kept for the investigation. Straiker says STAR Labs saw agents do something they were never authorized to do in 85% of successful attacks. That is the case for stopping the actor, not just the action.

It is the incident-response playbook security teams already know from endpoints. When a laptop is compromised, you don’t block one process and hope. You isolate the host. Straiker is applying that habit to agents. It also maps neatly to the EU AI Act’s demand that humans can step in or stop high-risk AI systems.

Why it is winning now. Straiker says run-rate revenue grew more than 15x in under a year, off an undisclosed base, and that its customers include Fortune 500 companies and frontier AI labs. Company claims, no named logos in the release. Gartner listed it as a sample vendor for agentic AI security in two October 2026 Hype Cycle reports. And its research gets noticed: in October it published an inside look at a criminal group running AI models in its attacks on Latin American banks.

The CyberMerge view. AI agent security is crowded. Discovery and prompt filtering are turning into features that Palo Alto, CrowdStrike, Zscaler and the cloud providers will bundle. The kill switch is a better wedge, because it speaks the SOC’s language: contain first, investigate second.

The risk is depth. A kill switch only works on agents Straiker can see and reach. Homegrown agents, coding agents on laptops and SaaS copilots all live in different places. And a 15x growth rate off an unknown base tells you direction, not size.

Bottom line. Straiker is betting that agents will be treated like employees with access, and that every employee eventually needs an off switch. Watch for named customers, kill-switch coverage across third-party agents, and whether the big platforms copy the feature or buy the company.

Sources