SECURITY

AI cyber warfare startup Twenty raises $100M at $1B valuation

Twenty Inc., a startup that develops cyber warfare software for the U.S. military and intelligence community, has raised $100 million in funding.

Accel led the Series B investment. Twenty stated in its announcement of the deal today that Friends & Family Capital, Point72 Ventures and Caffeinated Capital participated as well. The company is now valued at $1 billion.

Twenty was launched in 2024 by a group of former public sector cybersecurity experts. Several of its co-founders also held senior roles at Palo Alto Networks Inc. Twenty Chief Executive Joe Lin was a vice president of product management at the cybersecurity giant.

Twenty’s cyber warfare software uses artificial intelligence to reduce manual work for users. According to the company, some of the tasks that its systems automate can take humans weeks to complete. At the same time, Twenty says the systems “keep human judgment at the center” by facilitating manual evaluations.

AI models are proving increasingly adept at launching cyberattacks. In April, Anthropic PBC debuted Claude Mythos Preview, a large language model that it says has discovered more than 23,000 software vulnerabilities. The company estimates that over a quarter of those flaws are high-severity or critical.

Notably, Mythos Preview can not only identify vulnerabilities but also exploit them. During one internal test, Anthropic deployed an agent powered by the model in a virtual machine and asked it to compromise Firefox. The agent generated more than a dozen proof-of-concept cyberattacks in under an hour.

This month, Anthropic introduced a new frontier model called Claude Mythos 5. The company says it’s even better than Mythos Preview at cybersecurity research tasks. Mythos 5 topped its predecessor’s score on SWE-Bench, a widely used AI programming benchmark, by 2.5%.

Twenty hasn’t shared any information about the AI technology that powers its software. The company has raised a total of $138 million to date, which means that it’s likely not developing a general-purpose frontier model such as Mythos 5. Such algorithms require billions of dollars worth of hardware to develop.

AI models with offensive cybersecurity capabilities are useful for not only cyber warfare but also red teaming. That’s the process of searching for vulnerabilities in a company’s infrastructure by simulating hacking attempts. Furthermore, cybersecurity teams can use simulated breaches to practice incident response procedures. 

“We are building the industrial base for American cyber power: the AI-enabled capabilities our warfighters need to disrupt threats at their origin,” Lin said.

Twenty will use the proceeds from its funding round to finance research and development initiatives.

Photo: Unsplash