Before AI Agents Get a Job, AIUC Wants to Test Their Safety
AI agent
Software that can complete several steps toward a goal.
AIUC-1
A safety standard created by AIUC for checking AI agents.
SOC 2
A cybersecurity standard that AIUC used as a model.
What happened
Artificial Intelligence Underwriting Company, or AIUC, announced a $40 million Series A round on September 15. TechCrunch reported the funding and testing plan. Ribbit Capital led the round. First Harmonic also participated. AIUC had already raised $15 million in seed funding. Its reported total is now $55 million.
The founders are Rune Kvist and Rajiv Dattani. Kvist was an early Anthropic employee. Dattani served as METR's chief operating officer from 2024 to 2025. He remains on METR's board. AIUC names Cursor, Lovable, Harvey, and ElevenLabs as customers.
How the service works
An AI agent is a system that can take a goal and complete several steps. AIUC wants companies to test such systems before giving them important work.
The company created AIUC-1. It used SOC 2, a widely used cybersecurity standard, as a model. AIUC first asked about 250 security and risk leaders what buyers should check. Their answers shaped the standard and its tests.
AIUC says it runs about 5,000 tests. The tests examine responses to unsafe instructions, wrong answers, and possible data leaks. AI agents help run the tests. AI also helps analyze the results. Humans verify the final audit.
Customers receive a report of about 100 pages. It describes where an agent performed safely and reliably. It also marks areas that need caution. AIUC is not proposing to sit inside customer companies. It offers an outside assessment before deployment.
Why it matters
Businesses do not need only a smart model. They need a system that respects promises to customers. A bank, hospital, or government team may need to know what an agent will and will not do. As agents handle more steps, that promise becomes harder to verify in advance.
This creates a market for independent evidence. AIUC's idea resembles a familiar security audit, but applies it to agent behavior. METR has also tested advanced AI. Its recent focus, however, has mainly been task performance. AIUC emphasizes safety checks for enterprise buyers.
What we can confirm
The funding, AIUC-1, the testing service, and the human review process come from AIUC's announcement and descriptions reported by TechCrunch. The reported facts show a company building a safety assessment business. They do not show that every tested agent is safe.
What we still do not know
A five-thousand-test package is still a snapshot. We do not know how well its results predict months of real workplace use. We also do not know whether AIUC-1 will become a common industry standard. Public evidence about customer outcomes, later incidents, or measurable improvements is still limited.
What to watch next
The important signals will be adoption and independent comparison. Will more companies request AIUC-1 reviews? Will other evaluators reach similar conclusions? Will customers report fewer problems after testing? Those answers will show whether AI safety checks become a normal buying step, rather than a promise on paper.
A New Safety Check for AI Agents
📰 Full story: Before AI Agents Get a Job, AIUC Wants to Test Their Safety
AIUC checks AI agents before companies trust them. [TechCrunch reported](https://techcrunch.com/2026/09/15/early-anthropic-hire-former-metr-coo-have-found-a-way-to-rein-in-rogue-ai-agents/) that it raised $40 million.
AI agent
Software that can complete several steps toward a goal.
AIUC-1
A standard for checking how safely an AI agent behaves.
SOC 2
A cybersecurity standard used as a model by AIUC.
💡 The gist
- AIUC checks AI agents before companies trust them.
- It uses about 5,000 tests and human review.
- The results show risks, not perfect safety.
Artificial Intelligence Underwriting Company (AIUC) is a safety-testing company. Rune Kvist helped start it. He worked early at Anthropic, an AI company. Rajiv Dattani helped start it too. He was METR's former chief operating officer. METR is an AI safety research group.
AIUC raised $40 million in a new funding round. Ribbit Capital led it. First Harmonic also joined. AIUC previously raised $15 million. That makes $55 million in total. AIUC names Cursor, Lovable, Harvey, and ElevenLabs as customers.
An AI agent can complete several steps toward a goal. Companies want agents to work independently. They also need agents to respect clear limits. AIUC created AIUC-1 for this purpose. It used SOC 2 as a model. About 250 security and risk leaders helped shape its questions.
AIUC runs about 5,000 tests. The tests look for wrong answers and information leaks. They also check responses to unsafe instructions. AI agents help run the tests. AI helps study the results too. Humans review the final audit. Customers receive a report of about 100 pages. It shows reliable areas and remaining concerns.
This matters because smarter AI can be harder to control. Businesses make promises to customers. They need evidence before giving agents important work. An outside review can reveal problems earlier. It can also help buyers compare different systems.
Still, a test is not a lifetime guarantee. We do not know if AIUC-1 will spread widely. We also lack public results from long-term use. The next test is real-world experience. Watch for independent comparisons and reported improvements.
A Company Gives AI a Safety Check
📰 Full story: Before AI Agents Get a Job, AIUC Wants to Test Their Safety
AIUC tests AI before a business gives it important work. [TechCrunch reported](https://techcrunch.com/2026/09/15/early-anthropic-hire-former-metr-coo-have-found-a-way-to-rein-in-rogue-ai-agents/) its new funding.
AI agent
An AI worker that can do several steps for a job.
METR
A group that studies AI safety.
AIUC
A company that checks whether AI is safe to use.
Anthropic is an AI company. Rune Kvist worked there early. Rajiv Dattani led operations at METR, an AI safety group. They started AIUC.
AIUC checks an AI worker before people trust it. The AI worker is called an AI agent. It can do several steps for a job.
AIUC gives it about 5,000 tests. The tests look for mistakes and secret leaks. AI helps with the tests. People check the final results. The report is about 100 pages.
AIUC raised $40 million. It had already raised $15 million. $40 million + $15 million = $55 million.
This is like a safety check before a trip. It can show problems. It cannot promise that every trip is safe. AIUC hopes its checks help businesses choose carefully.