Plurai raises $13M seed to stress-test AI agents before launch
What's the deal? PluraiDealroom has a profile for this one. Try Dealroom → has raised $13 million in seed funding to expand a platform that simulates corporate environments and stress-tests autonomous AI agents before they go live. Team8 led the round, with participation from U&IDealroom has a profile for this one. Try Dealroom →, Velocity, MercerDealroom has a profile for this one. Try Dealroom →, and angel investors including CyberArk founder Udi Mokady and Google DeepMind researcher Reah Miyara.
What's the endgame? Founded by former Nexar researchers Ilan Kadar and Elad Levi, the New York-based startup borrows an approach from autonomous driving: simulation. It creates a digital twin of an organisation's tools, knowledge bases, and policies, then runs thousands of synthetic interactions—including adversarial ones—to flag hallucinations, policy breaches, and failed tasks before real users hit them.
What else? Plurai uses simulation data to post-train small language models that act as always-on monitors in production, a cheaper alternative to inspecting every interaction with a frontier model. In work with Nvidia and Nebius for an enterprise customer, it says its post-trained models beat frontier models from OpenAI and Anthropic on accuracy at roughly 40 to 44 times lower cost.
Why now? Building an agent is easy; making it reliable across unpredictable interactions is not. Plurai also opened general availability of its platform, previously limited to select partners, adding a free tier for developers. It employs about 12 people across Israel and the US and plans to double headcount by the end of 2026.
By the numbers: The round sits in the top 5% of seed rounds in the US over the trailing 48 months, a sample of 17,141 deals. Plurai's own claims are bolder still: design partners reportedly cut failures reaching users by 82% and operating costs by as much as 97%—figures drawn from early customer work that await independent validation.
The signal: The round lands in a fast-growing market for AI-agent evaluation and observability, where firms raised $272 million through September 2026, up from $166 million a year earlier. As enterprises deploy agents at scale, the bet is shifting from monitoring failures after deployment to simulating them before—an approach that, if it holds up, could define reliability tooling for the agent era.
Read more: TechCrunch
Image credit: Official GDC
