Startup Reinvents Cybersecurity To Stop AI Agents From Being Hacked

As AI agents increasingly handle sensitive data, cybersecurity needs a complete redesign. Multifactor is stepping up with open-source solutions to stop hackers from exploiting AI’s weakest links.

Image Credit: 3rdtimeluckystudio / Shutterstock

Cybersecurity for the AI Era

Hertz Fellow Vivek Nair thinks that traditional cybersecurity approaches don't cut it when it comes to the world of rapidly advancing AI systems. His startup company, Multifactor, aims to provide new security solutions designed from the ground up to protect consumers and businesses from AI-related data breaches.

Support from the Hertz Foundation

Nair has now won funds from the Hertz Foundation via the Harold Newman and David Galas Entrepreneurial Initiative, which will be used to support the development of multiple new cybersecurity products aimed at both companies and individual consumers. Multifactor has also been selected for the fall 2025 batch of Y Combinator, a prestigious and competitive startup accelerator that has helped launch top companies, including Airbnb, Instacart, Reddit, and Stripe.

"We want to specifically understand the vulnerabilities that are caused by the adoption of AI and AI agents in different applications, and then solve those vulnerabilities," says Nair. "The Hertz Community has been key for Multifactor's trajectory, and this new support helps us keep developing our open-source technologies."

The Entrepreneurial Initiative

Since 2012, the Harold Newman and David Galas Entrepreneurial Initiative has provided financial and professional support to Hertz Fellows who propose innovative entrepreneurial projects. A panel of experts judges the projects, and one or two are selected each year to receive support. The Initiative recognizes honorary Hertz Fellow Harold Newman, a former member of the Hertz Foundation board of directors, as well as David Galas, a Hertz Fellow and chairman of the board. For each selected Hertz Fellow, the Initiative provides up to $25,000 in funding, as well as mentoring and feedback from others within the Hertz Community.

Nair’s Path to Cybersecurity

Nair has been studying cybersecurity since high school, when he launched several startups and began working for a healthcare tech startup in Singapore to address the issue of keeping patients' sensitive health data safe. As a Hertz Fellow at UC Berkeley, he later completed a PhD in computer science and developed new cryptographic methods. He also worked for the CIA, applying his expertise to national cybersecurity challenges.

AI Agent Vulnerabilities

Today, he is focused on how AI agents open new windows into people's data. He points toward a recent demonstration by security researchers showing how to hijack Google's AI bot, Gemini, with a poisoned Google Calendar invitation. Even if a user doesn't open the calendar invitation, Gemini accesses it. It follows its instructions on controlling devices associated with Google, potentially leading to breaches such as unlocking linked doors in someone's home.

"Instead of tricking a person to do something they're not supposed to, like open an email, you're tricking an AI agent," Nair explains.

Multifactor’s Approach

Multifactor, co-founded by Nair and mathematician and cryptographer Colin Roberts, has already developed cryptographic techniques that protect against these vulnerabilities. Their products enable AI agents to log into systems and accounts without ever storing credentials in plain text, ensuring that an AI agent has only the authorization it needs for the exact tools and features it requires. They have also developed a method to collect extremely detailed logs of the actions an AI agent carries out.

"Right now, a lot of companies are trying to fit the conventional human security model to AI agents," says Nair. "They are forcing AI agents to interact with systems in the same way a person does, but this leads to a lot of vulnerabilities. We're building solutions that enable the exact same functionality, just without the security risk."

Adoption and Future Plans

Already, Multifactor has partnered with several large companies that want to ensure their AI workflows are secured. They are also unveiling a waitlist for a more consumer-oriented security platform. The technology, acting like an advanced password manager, would allow people to grant AI agents, such as ChatGPT or Claude, access to specific data from other accounts without requiring direct access to passwords.

"This is a cool way that we hope everyday users of AI can take advantage of our platform to make their day-to-day lives easier and safer," says Nair.

Alignment with the Hertz Mission

Nair says that the mission of Multifactor, a public benefit company making most of its technologies open-source, aligns closely with the Hertz Foundation's mission. Many of his early angel investors and backers have been Hertz Fellows, and his connections through Hertz have significantly influenced the development of his ideas into Multifactor. Hertz Fellows, including Philip Welkhoff and his wife Martina, Judy Savitskaya, and Max Kleiman-Weiner, have been integral to his success, he says.

The Next Pillar of Cybersecurity

"Cybersecurity is often broken down into pillars, identity security, web security, and network security. Sometimes you see technology creating new pillars, like the emergence of cloud security. We think agentic AI security is the next pillar of cybersecurity and we want to be the company that defines it," Nair says. "Readers can be amongst the first to experience this technology for themselves by joining the waitlist at multifactor.com."

Comments

The opinions expressed here are the views of the writer and do not necessarily reflect the views and opinions of AZoAi.
Post a new comment
Post

Sign in to keep reading

We're committed to providing free access to quality science. By registering and providing insight into your preferences you're joining a community of over 1m science interested individuals and help us to provide you with insightful content whilst keeping our service free.

or

While we only use edited and approved content for Azthena answers, it may on occasions provide incorrect responses. Please confirm any data provided with the related suppliers or authors. We do not provide medical advice, if you search for medical information you must always consult a medical professional before acting on any information provided.

Your questions, but not your email details will be shared with OpenAI and retained for 30 days in accordance with their privacy principles.

Please do not ask questions that use sensitive or confidential information.

Read the full Terms & Conditions.