Superpower Daily

Superpower Daily

Your reading space

Sign in
OpenArt Launches Creative AI Arena That Splits Model Rankings b...Apple Releases macOS 27 With a Siri That Can Read Your ScreenSalesforce Connects Gemini Enterprise to Its Data Without Custo...Bolt’s New Policies Permit Training Beyond Its Forge-Only PitchMozilla Finds Chinese Open Models Are Within 4.4 Months of U.S....EUCLYD Raises More Than €200 Million for Non-GPU AI Inference S...OpenArt Launches Creative AI Arena That Splits Model Rankings b...Apple Releases macOS 27 With a Siri That Can Read Your ScreenSalesforce Connects Gemini Enterprise to Its Data Without Custo...Bolt’s New Policies Permit Training Beyond Its Forge-Only PitchMozilla Finds Chinese Open Models Are Within 4.4 Months of U.S....EUCLYD Raises More Than €200 Million for Non-GPU AI Inference S...
Swipe

AIUC Raises $40 Million to Audit Enterprise AI Agents

The startup is trying to give enterprise buyers a clearer view of how an AI agent behaves when faced with failures such as jailbreaks, hallucinations and data leaks.

By 2 min read
AIUC Raises $40 Million to Audit Enterprise AI Agents
AIUC Raises $40 Million to Audit Enterprise AI Agents

Listen to this story

The audio brief

About 1:34
0:001:34
Read transcript
AIUC has raised forty million dollars in a Series A led by Ribbit Capital, bringing the company’s disclosed funding to fifty-five million. Its pitch is straightforward: before an enterprise gives an AI agent meaningful work or access to sensitive data, buyers should have an independent record of how that agent behaves when things go wrong. The company’s main product, AIUC-1, is modeled partly on SOC 2, the software industry’s familiar security standard. But instead of focusing broadly on a vendor’s controls, AIUC-1 tests agent-specific failure modes. AIUC says it runs roughly five thousand scenarios involving jailbreaks, hallucinations, and data leaks, then produces a report of about one hundred pages. The assessment is AI-assisted: agents run the tests and analyze results, while people verify the final audit. That human check is central to the product’s positioning. An AIUC certificate is not a promise that an agent will never fail. It is meant to show where the system has behaved reliably, where risks remain, and what a buyer should weigh before deployment. The testing questions were shaped by a consortium of roughly 250 security and risk leaders. AIUC counts Cursor, Lovable, Harvey, and ElevenLabs among its customers. Founded by former Anthropic and METR executives Rune Kvist and Rajiv Dattani, the company now faces the harder market test: whether this kind of outside risk report becomes a purchasing requirement for enterprise AI agents, rather than an optional check after deployment.

Story brief

3 key points

AIUC has secured $40 million to commercialize independent audits for enterprise AI agents, bringing its disclosed funding to $55 million. Its AIUC-1 service runs about 5,000 scenarios covering jailbreaks, hallucinations, and data leaks, then produces an approximately 100-page report. Agents and AI systems automate testing and analysis, but humans verify the final assessment. The key market question is whether buyers...

  1. 01

    Ribbit Capital led the Series A; First Harmonic also participated.

  2. 02

    AIUC-1 is modeled partly on SOC 2 but evaluates agent-specific failure modes.

  3. 03

    Named customers include Cursor, Lovable, Harvey, and ElevenLabs.

AIUC has raised a $40 million Series A for its effort to audit and certify AI agents used by enterprises. Its bet is that companies need a clearer record of an agent’s failure modes before they give it meaningful work or access to data.

Ribbit Capital led the round, with First Harmonic participating. AIUC previously raised a $15 million seed round, bringing its disclosed funding to $55 million. The company was founded by Rune Kvist, an early Anthropic employee, and Rajiv Dattani, METR’s former chief operating officer and a current board member.

A safety check for buying agents

AIUC has developed AIUC-1, a safety standard and testing service for enterprise agents. The company took inspiration from SOC 2, a widely used cybersecurity compliance standard for software vendors. AIUC’s goal is an independent assessment that tells a buyer where an agent performs safely and reliably, and where concerns remain.

What AIUC says it examines

  • Roughly 5,000 test scenarios involving jailbreaks, hallucinations and data leaks.
  • A roughly 100-page report that identifies safer, more reliable behavior as well as concerns.
  • Questions shaped by a consortium of about 250 security and risk leaders, according to AIUC.

Automation, with a human final check

AIUC uses AI agents to run its tests and analyze the results, while humans verify the final audit. The arrangement lets the company automate large test suites without presenting the final assessment as entirely machine-reviewed.

That distinction is important to the product’s pitch. A certificate is not a promise that an agent cannot fail; AIUC describes its reports as a way to show buyers where an agent can be trusted and where they should weigh remaining concerns before deciding to buy.

An early market test

AIUC names Cursor, Lovable, Harvey and ElevenLabs among its customers. The new funding gives it more capital to pursue a simple but demanding aim: make outside safety testing useful enough that enterprise buyers treat it as part of choosing an AI agent, rather than an optional extra after deployment.

Sources

  1. techcrunch.comEarly Anthropic hire, former METR COO have found a way to rein in rogue AI agents | TechCrunch

Loading discussion...

Related coverage

Continue reading