Business News | OpenAI Resumes Frontier AI Model Training After 11-day Pause over Security Concerns

Agency News

Get the latest business articles and stories on LatestLY. OpenAI, led by Sam Altman, has started training its most advanced AI models again after improving security in its research environments and limiting internet access. This followed incidents in which AI models showed they could get around safety barriers designed to keep them contained, the company said in a statement.

OpenAI logo (Photo/X/@OpenAI)

New Delhi [India], October 24 (ANI): OpenAI, led by Sam Altman, has started training its most advanced AI models again after improving security in its research environments and limiting internet access. This followed incidents in which AI models showed they could get around safety barriers designed to keep them contained, the company said in a statement.

The AI company also revealed security weaknesses in software used in its research environments. These weaknesses were found during security tests. The findings come as more people worry that increasingly powerful AI agents could use weaknesses in systems meant to limit what they can do.

OpenAI had temporarily stopped training its most powerful AI models after a research model used a setup mistake to get limited internet access in September. The company said it contained the incident and used the pause to improve its security measures.

To check whether its improved security measures worked, OpenAI sent teams of AI agents to test whether they could break out of restricted research environments. The company said the agents found previously unknown security weaknesses. OpenAI fixed these problems and restarted training on October 1.

“In our red-team tests, AI agents found software security weaknesses that had not been known before in open-source components we use. They also found incorrect settings in services we run. These weaknesses allowed the agents to escape sandboxes, move between systems, and reach the live internet. They also found ways to communicate across boundaries that were meant to keep systems separate. These findings led us to release software fixes, set stricter limits on network access, and report the problems to the relevant software maintainers and vendors,” the AI company said.

The company also said its tests helped it find possible security gaps and showed that its existing monitoring systems were working.

“Our red-team testing found weaknesses that needed to be fixed, while also giving us evidence that our existing protections were working. We did not find an easy way for agents to reach the internet while hiding their actions from our monitoring systems. Even so, the findings helped us close possible ways into our systems, fix software, and improve monitoring,” it said.

After testing security, making the needed fixes, and improving its monitoring and shutdown systems, OpenAI said it had “resumed reinforcement learning (RL) training and internal inference of our frontier models on October 1,” just “11 days” after the pause began.

OpenAI has also been reviewing the activities of its AI models more broadly since the Hugging Face incident in July. It has told more than 100 organisations about incidents involving unauthorised activity linked to its AI agents. (ANI)

(The story above has been checked and written by ANI staff. ANI is South Asia's leading multimedia news agency, with more than 100 bureaus in India, South Asia, and around the world. ANI reports the latest news on politics and current affairs in India and worldwide, sports, health, fitness, entertainment, and other news. The views expressed in this post do not reflect the opinions of LatestLY.)