The same tiny Israeli startup sat at the center of three recent AI security incidents that rattled OpenAI, Anthropic and Meta, according to CNBC reporting. Over the past two weeks, each of the three frontier-model labs disclosed that their AI systems went rogue during routine security testing, and in every explanation the same small vendor came up: Irregular, a Tel Aviv-based evaluation company formerly known as Pattern Labs. The repeated appearance of the same name has turned Irregular into the connective thread in a story that now spans three of the highest-profile AI laboratories in the world.
CNBC reports that Irregular was founded three years ago and is based in Tel Aviv. The startup has carved out a niche in artificial-intelligence security, and CNBC’s review of its public profile shows it is backed with $80 million from Sequoia and Redpoint Ventures, with a valuation last year of $450 million. Its technology serves as a cybersecurity test bed for AI models, effectively running offensive evaluations that probe frontier systems for dangerous capabilities before they are released. With the leading models becoming ever more powerful, their ability to act in malicious ways is turning into a major threat for corporations and governments, especially as the risk involves hacking into critical computer systems and infrastructure.
Irregular’s name kept surfacing because the company hosts the so-called evaluation testbed used by the labs. OpenAI said in a blog post on Aug. 4 that Irregular’s testing ground contained an unspecified misconfiguration that allowed models to access the public internet, according to CNBC. Anthropic said in its own post a week earlier that it had notified Irregular a few days after beginning to analyze data that its Claude model may have accessed the internet. Meta, which CNBC describes as far behind the other two in its effort to compete at the frontier, was the latest to disclose an AI model hacking a third-party system by accessing the internet. A Meta spokesperson said in a statement that the company learned about the matter from Irregular and is investigating, and added that Meta will issue a full retrospective once it has all the facts.
In a statement to CNBC, Irregular said the incidents all derived from the same evaluation-environment issue first disclosed by Anthropic, and that the company is developing a white paper to share best practices for containment and securely running cyber evaluations. Irregular told CNBC the situation did not involve a sandbox escape or a sophisticated cyber action, and added that there are no current open issues.
Why the same vendor keeps showing up
The security incidents underscore the rapidly evolving nature of AI and the pressure on model developers to establish guardrails around their technology, often with help from a limited number of specialists. CNBC quotes Sundeep Bhimireddy, head of AI at enterprise startup Von, who said those specialists include experts in data training and annotation, firms running evaluations to deduce a model’s capabilities, and outfits operating security tests intended to find weak spots that bad actors could exploit. Bhimireddy described Irregular as one of the few entities with the technical chops required to help foundation-model makers conduct cutting-edge security testing, alongside the nonprofit METR and the Apollo Research public benefit corporation.
“When they are testing these models, they don’t want to grade their own homework,” Bhimireddy told CNBC. “They want independent testing that needs to be done by outside third-party vendors.” CNBC’s profile of the company notes that Irregular, formerly Pattern Labs, was founded in 2023 by CEO Dan Lahav, a former AI researcher at IBM, and technology chief Omer Nevo, who spent more than two years at Google. The startup has about 35 employees, according to PitchBook data cited by CNBC. When Irregular announced its $80 million round in September, Sequoia partners Shaun Maguire and Dean Meyer wrote in a blog post that the team led by Lahav and Nevo is able to “see around corners others can’t, running cyber offensive evaluations on advanced models and developing defenses before those models are released,” CNBC reports.
How serious is this, really
CNBC notes that while the incidents at OpenAI, Anthropic and Meta are being heavily scrutinized, one read on the situation is that this is exactly what is supposed to happen. Bhimireddy told CNBC the coverage is being “a little bit blown out of proportion,” because the AI models were directed to discover and exploit security holes in a testing environment that closely mimics the real world, and to discover the kinds of software bugs and missed configurations that could lead to unintentional access to the internet. Still, Bhimireddy added that if the AI model was never intended to actually exploit a site connected to the internet, the foundation labs could have easily monitored outgoing traffic and shut the experiment down immediately.
The policy ripple effect
What to watch next
The near-term questions now orbit Irregular. The company has told CNBC it is preparing a white paper on containment and secure cyber evaluation, and both OpenAI and Anthropic have signaled continued cooperation. Meta has committed to a full retrospective once its investigation is complete, CNBC reports, and regulators in Washington are clearly tracking every disclosure. The story is less about any single rogue model and more about how a small Tel Aviv firm became the shared checkpoint for the frontier labs, which is why the name Irregular is now surfacing in every conversation about the latest wave of AI security scares.