Several of the world’s leading AI companies are warning that their systems have begun to “break free” from the closed environments meant to contain them. Is this the start of a genuine security crisis — or a PR stunt aimed at investors?
“It is not the technology that has lost control, it is the companies that are acting irresponsibly,” says Virginia Dignum, professor of responsible AI at Umeå University, who argues the danger is real but widely misdiagnosed.
New technology is almost always tested before launch, and AI is no exception; major players including OpenAI, Anthropic, DeepSeek and Moonshot run various forms of safety testing before releasing a new version.
The alarm sharpened on July 21, when OpenAI disclosed that a model it had trained had “escaped” its test environment — a claim that has intensified calls to slow development until oversight catches up.