Probably, a startup working on confidence in artificial intelligence, has raised $9 million in funding, according to TechCrunch on June 16, 2026.

The $9 million funding round places Probably alongside other AI startups that closed similarly sized rounds recently. Sprouts.ai raised $9 million in May 2026 from True Global Ventures and Accel. GRAI pulled in $9 million in April 2026, per Vestbee. And Worktrace AI, founded by former OpenAI employee Angela Jiang, launched with $9 million last year.

What Is Probably Building

Probably is solving a straightforward but significant problem. When you ask an AI system a question, it gives you an answer with perfect confidence — but that doesn't mean the answer is right. The AI might be accurately recalling something it learned during training, or it might be making something up entirely. To you, both answers sound equally certain. That is the core issue.

What Probably is building is a way for AI systems to say how confident they are. Instead of just answering your question, the system would say something like: "I'm 85 percent confident in this answer" or "I'm only 20 percent confident." That signal is incredibly useful.

Where does this matter most. Consider a lawyer using AI to review a contract. If the AI gives a legal interpretation with complete confidence, the lawyer might trust it. But if the AI could say "I'm only 40 percent sure about this clause," the lawyer would know to double-check manually. That is a practical difference. The same applies to doctors using AI to help diagnose patients, or financial analysts asking AI for market insights. In all these cases, knowing the AI's confidence level is as important as the answer itself.

Why This Problem Is Bigger Than It Sounds

AI research scientists have studied confidence and uncertainty in AI systems for years. But most teams actually building AI products don't use those techniques. Instead, they find workarounds. They might feed the AI relevant documents alongside the question so it stays grounded. They might ask the AI to explain its reasoning step by step. Or they might have a human review the answer before it is used. All of these help, but none directly tell you: "how much should you trust this."

As AI moves into fields with real stakes — healthcare, law, finance, insurance — the need for confidence becomes pressing. Regulators and companies both want to know: can you audit this system. Does it make errors consistently. Can you trust it. Current AI systems cannot answer these questions directly.

Probably's bet is that confidence tooling will become essential infrastructure, especially in regulated industries. The $9 million raise suggests investors agree. Capital is still flowing into early-stage AI infrastructure, which indicates investors believe there are still basic problems to solve as AI moves from research labs into real-world use.

One realistic caveat: building reliable confidence scoring is genuinely difficult. If Probably's tools attach confidence signals to AI answers after the fact — rather than building confidence into the AI system itself — then the quality of those signals depends entirely on the training examples Probably uses. Bad training examples would produce misleading confidence scores. That is something to watch as the company grows and deploys at scale.

Whether confidence tooling stays as a separate product or eventually gets built into the major AI platforms — the companies that make the AI models themselves — is an open question. That pattern has played out before in software history, but it does not always work out well for the startups that pioneered the category.