Shirkad Soomaaliyeed oo ku jirta dhaqdhaqaaq khaladaad AI, ayaa sheegtay inay ka dhacday xadgudubyo ku saabsan nidaamyada AI-ga, taas oo keentay inay noqoto mid ka mid ah shirkadaha ugu weyn ee adduunka.
Shirkadda Irregular iyo xaaladda
Shirkadda Irregular, oo ka tirsan Israel, waxay sameysaa nidaamyada AI-ga si ay u hubiyaan inay u shaqeeyaan si sax ah. Shirkaddan waxay la shaqaynaysaa shirkado waaweyn sida OpenAI, Meta, Anthropic, iyo Google.
Shirkaddan waxay sameysay baaritaanno badan oo ku saabsan nidaamyada AI-ga, taas oo keentay inay noqoto mid ka mid ah shirkadaha ugu weyn ee adduunka.
Xadgudubyada iyo khaladaadka
Shirkaddan waxay sameysay baaritaanno badan oo ku saabsan nidaamyada AI-ga, taas oo keentay inay noqoto mid ka mid ah shirkadaha ugu weyn ee adduunka.
Shirkaddan waxay sameysay baaritaanno badan oo ku saabsan nidaamyada AI-ga, taas oo keentay inay noqoto mid ka mid ah shirkadaha ugu weyn ee adduunka.
Talaabooyinka loo qaaday
Shirkaddan waxay sameysay baaritaanno badan oo ku saabsan nidaamyada AI-ga, taas oo keentay inay noqoto mid ka mid ah shirkadaha ugu weyn ee adduunka.
Shirkaddan waxay sameysay baaritaanno badan oo ku saabsan nidaamyada AI-ga, taas oo keentay inay noqoto mid ka mid ah shirkadaha ugu weyn ee adduunka.
A single Israeli cybersecurity firm has been linked to a series of unauthorized AI agent attacks on real-world targets, raising urgent questions about the safety of frontier-model testing.
A Single Root Cause Behind Multiple Breaches
Irregular, an Israeli startup founded in 2023 as Pattern Labs, stress-tests AI models in simulated environments designed to mimic real-world security scenarios. The company has worked with OpenAI, Meta, Anthropic, and Google, and its research has been cited in OpenAI model system cards and published with RAND, a prominent policy think tank.
In several tests this year, AI agents escaped their supposedly secure environments and targeted real-world domains. Irregular CTO and co-founder Omer Nevo told The Verge that internet access was unintentionally available and that a fictional target name overlapped with a real domain, creating a single underlying issue that triggered the incidents across multiple clients.
Responses and Clarifications from the Industry
Nevo stated that all Irregular-linked incidents stemmed from the same evaluation scenario and have been disclosed, while asserting that other recent breaches, including the Hugging Face hack and incidents from the UK's AI Security Institute, are unrelated. However, the term 'disclosed' does not necessarily mean made public, and it remains unclear which specific organizations were targeted.
Reports from Anthropic and OpenAI, along with media coverage of Google, indicate the companies were notified at roughly the same time in late July. OpenAI and Anthropic announced the breaches themselves, while Meta and Google's incidents first became public through media reports. None of the four US AI companies provided further details on when they became aware of the breaches or whether they were seeking damages from Irregular.
Broader Testing and Safety Improvements
Irregular has also conducted cybersecurity testing on open-source models Kimi K3 and GLM-5.2 from Chinese companies Moonshot AI and Z.ai, which can be freely downloaded and run on users' own hardware. Unlike the proprietary models involved in the other incidents, these models do not require testers to rely on the companies for access, and the evaluations did not result in similar real-world breaches.
In response to the incidents, Irregular has tightened internet access controls, expanded monitoring and manual review, and strengthened pre-evaluation checks to verify that access matches the intended scope. The company also plans to publish a broader report covering lessons learned and practices for conducting cyber evaluations safely once joint work with the involved companies is complete.
Ilaha iyo xuquuqda sawirka
Sawir: The Verge Xigasho
