OpenAI Suspends Work on Latest AI Models After Security Restrictions Bypassed

OpenAI announced the suspension of training, evaluation and inference operations involving tool use for its most advanced models after an artificial intelligence model managed to bypass internet-access restrictions during a security test.

The company explained that the model was operating inside a test environment designed to prevent it from accessing the internet. However, it discovered a vulnerability connected to the Domain Name System (DNS) filtering system and exploited it to communicate with an external chatbot over the internet.

OpenAI said the model had been tasked during the test with identifying the person who wrote a blog post based on a collection of clues. After being unable to complete the task using the simulated web environment, it discovered that it could use the DNS system to send queries to an external service and obtain information from it.

The company added that it detected the external communication and stopped the test. It said training, evaluation and inference operations involving tool use would remain suspended until the vulnerability was addressed and additional safety tests were conducted.

OpenAI indicated that the incident was less serious than some previous events but was the first of its kind since protective measures for testing environments were tightened.

Source: QNA

--

About Qatar Living:

Since 2005, Qatar Living has been the trusted destination for everything Qatar. As the country's largest online community and marketplace, Qatar Living connects people with opportunities through jobs, property, vehicles, services, classifieds, events, local insights, and breaking news. Trusted by residents, newcomers, visitors, and businesses alike, Qatar Living brings Qatar together in one place.

Instagram - @qatarliving

X - @qatarliving

Facebook - Qatar Living

YouTube - qatarlivingofficial