Skip to content
Rush Commerce
AI & Automation3 min read

Harvey raises $550M and buys Guardrails AI for agent evals

Harvey raised $550M at $15.5B, bought an agent-eval company the same day, and built its own model on open weights. The moat moved off the model.

Harvey announced a $550 million round at a $15.5 billion valuation on September 9. Same day, same company, second announcement: it acquired Guardrails AI, an agent security platform — Harvey's fourth acquisition of 2026. Put those next to what Harvey has been shipping and you get the clearest signal yet about where value actually sits in vertical AI. It is not the model.

What actually happened

The round was co-led by Diffusion and Lightspeed, with Sapphire Ventures and Whale Rock joining as new investors and Sequoia, Kleiner Perkins, a16z, Coatue, GIC and Goldman Sachs Alternatives following on, per Harvey's own announcement. Harvey says 80% of Am Law 100 firms and five of the Fortune 10 in-house legal teams are on the platform. TechCrunch puts total raised above $1.55 billion across at least eight priced rounds since 2023, and charts the climb: $8B in December 2025, $11B in March, $15.5B now. (Bloomberg headlined the valuation at $15.6B; we're using the number in Harvey's own post.)

Guardrails AI built the first open-source AI guardrails library and Snowglobe, a simulation environment for stress-testing agent behavior before it touches production. Co-founders Shreya Rajpal and Zayd Simjee join Harvey's product and engineering org. Harvey CEO Winston Weinberg framed the buy around one question: "Every firm we work with asks the same question before they let an agent near real client work: how do you know what it will do?"

The third piece completes the picture. Harvey's own model, Harvey Tenet, is post-trained from the open-weight Kimi K3 — and Harvey is telling customers to post-train their own open-weight models too.

Why it matters for your business

Read the three together. The best-funded vertical AI company on the planet raised half a billion dollars and did not spend it on a frontier model. It built on someone else's open weights, and it bought an agent evaluation team. That is a company saying out loud that the model is a commodity input and the answer to "what will it do?" is the product.

That reframes your problem, because the blocker on shipping an agent in a 12-person business is the same blocker as in an Am Law 100 firm. It was never model quality. It's that nobody can say what the thing will do on a Tuesday with a weird invoice.

At your scale the same three parts are cheap. A base model behind a gateway so you can swap it. Your own domain data — the closed tickets, the past quotes, the actual email threads. And a written spec of what the agent may and may not do, backed by a regression suite that replays 50 real cases and diffs the output before anything deploys. That last part is the one almost nobody builds. If you can't replay real history against a changed prompt and see what moved, you don't have an agent in production. You have a demo you're hoping about.

Key takeaways

  • Harvey raised $550M at a $15.5B valuation on September 9, co-led by Diffusion and Lightspeed — up from $11B in March and $8B in December 2025
  • Same-day acquisition of Guardrails AI (agent simulation and evals) is Harvey's fourth of 2026 — it bought an eval harness, not a model
  • Harvey Tenet is post-trained from open-weight Kimi K3, and Harvey is pushing customers to post-train their own open-weight models
  • The deployment blocker is "how do you know what it will do?" — not model quality
  • Your version costs almost nothing: swappable base model, your own data, and a replay suite of 50 real cases that runs before every deploy

Can you prove what your agent will do before it emails a customer? We build agent systems on a swappable model layer with a replay suite over your real history, so a prompt change is a diff you can read instead of a risk you absorb. See how we build it, or bring us your riskiest workflow.

Sources: Harvey funding announcement, Guardrails AI joins Harvey, TechCrunch.

  • #legal-tech
  • #ai-agents
  • #agent-evals
  • #open-weights
  • #vendor-risk
TR

Tommy Rush — Founder, Rush Commerce

Operator turned builder. 15+ years running operations — now shipping the systems businesses run on. More

Get The Rush Report weekly — one email, zero fluff.