OpenAI announced that it will stop training its frontier AI models on the live internet. The company said the decision follows “significant risks” linked to data exposure and cyber incidents. The change will affect all future training cycles, according to a statement released on Tuesday. This follows a recent security overhaul that uncovered a flaw in widely used software, which the company disclosed to regulators.
In London, the UK government outlined a new £800 million programme to strengthen AI security across public and private sectors. The funding will support research, policy development and the deployment of robust security frameworks. The announcement came after a series of cyber attacks that highlighted vulnerabilities in AI systems. The Treasury spokesperson confirmed the allocation in a briefing held on the same day OpenAI announced its training shift.
Nada de registros.
OpenAI’s move comes after a high‑profile breach that exposed internal code and training data. The company’s spokesperson said that training without internet access reduces the risk of leaking proprietary information. The shift also limits the model’s exposure to unverified online content, a concern that has grown as AI systems become more integrated into critical services.
The UK’s investment aligns with similar initiatives in the United States and Europe, where governments are allocating billions to secure AI infrastructure. The Treasury noted that the programme will help firms comply with emerging regulations and protect against ransomware and data theft. Industry experts say the funding could accelerate the development of secure AI platforms that meet stringent compliance standards.
Companies that rely on OpenAI’s models may need to adjust their data pipelines. The lack of internet connectivity could affect real‑time updates and the ability to incorporate fresh information. Meanwhile, the UK’s funding may encourage startups to adopt more secure architectures, potentially reshaping the competitive landscape. Both developments underscore the growing importance of cyber resilience in the AI sector.
