OpenAI has sharply tightened both physical and digital security after concluding that foreign actors are attempting to steal its proprietary artificial-intelligence technology, according to the Financial Times and other specialist outlets. The San Francisco company has installed fingerprint scanners at sensitive office areas, placed its most valuable computer systems offline by default and adopted a “deny-by-default” internet policy that blocks external connections unless specifically approved. It has also introduced strict “information tenting” protocols that limit staff access to individual projects; during development of its o1 model, for instance, only employees cleared for the so-called “Strawberry” tent could discuss the work even in communal spaces. Management has expanded background checks and bolstered its security team with former military and intelligence experts, including ex-Palantir chief information-security officer Dane Stuckey and retired U.S. Army general Paul Nakasone, who joined the board last year. People familiar with the company say the measures accelerated after Chinese start-up DeepSeek released a rival model in January that OpenAI believes was built using distillation techniques that replicate its technology. The clamp-down underscores escalating concerns in Washington and Silicon Valley that sophisticated AI systems are becoming targets for corporate and state-sponsored espionage.