Brief IA
›

AIR secures $50 million to safeguard AI agents

Business & Startups·Tom Levy·

AIR secures $50 million to safeguard AI agents

AIR secures $50 million to safeguard AI agents
Key Takeaways
1AIR raises $50 million in two rounds led by Sequoia and Greenoaks
2The platform discovers agents, intercepts their actions, and verifies tools and extensions through a whitelist
3AIR claims to filter about 27% of extensions and skills found online and has over 20 clients, including in finance and pharma
Why it matters : The increasing openness of enterprise systems to AI agents creates new security risks, which players like AIR are seeking to address with specialized solutions.

The security of AI agents is attracting significant investment. AIR, a startup founded by former members of Unit 8200, has announced $50 million to develop a platform that continuously evaluates the skills and extensions used by agents. The company claims to have over 20 clients and targets the most regulated sectors.

Massive rounds at rivals, AIR bets on re-verification

The AI agent security sector is drawing substantial funding: Zenity raised $125 million in a Series C round in August, and Noma secured $100 million in a Series B round last year. In this context, Yair Saban highlights AIR's ability to continuously assess the ecosystem of skills and extensions. Bogomil Balkansky, a partner at Sequoia, believes that the issue pertains to continuous re-verification rather than a simple scan, and he argues that real-time inspection and re-inspection at the scale of a fleet of agents are primarily infrastructure challenges. He notes that AIR has spent the past year building this pipeline. The company has around 40 employees, and the new capital will mainly be used to recruit researchers and expand its commercial efforts in the United States and Europe.

Discovery of agents, interception of actions, and whitelisting

AIR offers a platform that discovers active agents within the company and continuously evaluates the skills, tools, and components they use. It can prevent interactions with software or external sources that do not meet security criteria and includes a marketplace for verified extensions and skills. The system also identifies employees using unauthorized AI tools or personal accounts. An application layer is added to the agents to intercept and examine operations such as adding a skill or accessing content on the Internet. The tools and extensions employed by an agent are checked against a whitelist managed by the startup, which analyzes online offerings to detect potential modifications or malicious activities. A validated skill can become risky if a required package changes or if the developer's account is compromised. The platform currently filters about 27% of the extensions and skills found online.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Over 20 clients, with demand driven by finance and pharma

AIR claims to have more than 20 clients, with about a quarter being large enterprises. The strongest demand observed so far comes from highly regulated sectors, particularly financial services and the pharmaceutical industry.

Two close rounds for $50M, led by Sequoia and Greenoaks

AIR has raised a total of $50 million in two rounds completed within a few weeks of each other, according to Yair Saban. The first, led by Sequoia, amounts to $10 million; the second, led by Greenoaks, reaches $40 million. Investors such as Swish, Netz, Zach Frankel, Yinon Costica, Ofir Erlich, Anne Neuberger, Omer Adam, and Varun Anand also participated.

A risk of content poisoning drives the need to monitor tools

As companies open their systems more to AI agents, a supply chain made up of skills, plugins, MCP servers, and extensions is developing. AIR, a security startup, anticipates a specific need for monitoring this layer. Yair Saban compares the extensive use of agents to operating systems and notes that their tools are not regulated like drivers or applications, whereas the latter now require a signature, unlike in the early 2000s. He emphasizes the risk for agents that operate more autonomously, exposed to content poisoning rather than direct attacks.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.