Irregular collects $ 80 million to secure AI Frontier models

On Wednesday, the AI ​​security company announced $ 80 million in latest financing in the round run by Sequoia Capital and Redpoint Ventures, with the participation of the General Director of Wiz Assaf Rappaport. The source similar to the contract said that the round valued the $ 450 million irregular.

“We think that soon many economic activities will come from human interaction at AI and AI-Ai interaction,” said Techcrunch, co-founder of Dan Lahav: “And this will break a security pile along many points.”

Previously generally known as laboratory, irregular is already a significant player in AI’s assessment. The company’s works are cited in the security assessment for Claude 3.7 Sonnet and also Models O3 and O4-Mini OpenAI. Generally speaking, the company’s framework to assess the ability to detect the susceptibility of the model (Named solution) is widely used in the industry.

While irregularly carried out significant work on the existing risk of models, the company collects funds for something much more ambitious: detecting emerging risks and behaviors before they seem in the wild. The company has constructed a complicated system of simulated environments, enabling intensive testing of the model before its release.

“We have complex network simulations in which AI take the role of the attacker and the defender,” says co -founder Omer Nevo. “So when a new model appears, we can see where the defense persists and where not.”

Safety has turn out to be a point of intensive focus in the AI ​​industry, because the potential risks posed by border models have more risks. Opeli looked at his internal security measures this summer, taking into account potential corporate espionage.

At the same time, AI models are increasingly running in finding gaps in software security – power with serious implications for each attackers and defenders.

For irregular founders, this is the (*80*) of many safety pain caused by the growing capabilities of huge language models.

“If the goal of the Frontier laboratory is to create more and more sophisticated and talented models, our goal is to secure these models,” says Lahav. “But this is a movable goal, so there is many, much more work in the future by nature.”