OpenAI says frontier models will be trained in a fully air-gapped, closed environment

Cailian Press Financial News Agency

Open APP

Frontier AI model safety sparks concerns? OpenAI "declares resolve": will train fully offline!

Cailian Press Huang Junzhi Editor Ji Sheng 2026-10-24 10:13

① OpenAI is currently conducting frontier model training in a closed system with no internet access at all; ② The company officially restarted frontier model training on October 1 and is working with cloud service partners to advance the implementation of this offline training system; ③ The company emphasized that we must always have the ability to prevent frontier models from unauthorized, arbitrary access to the real internet.

Cailian Press, October 24 (Editor Huang Junzhi) OpenAI said on Friday that it will train its frontier artificial intelligence (AI) models in a safer environment, placing them in a closed system without internet access.

OpenAI confirmed in an official statement that it is currently conducting frontier model training in a closed system with no internet access at all. The company has specifically built a fully isolated environment called "offline world" for model training.

The company wrote in the statement: "We anticipate that future models will have the ability to autonomously identify and exploit vulnerabilities that humans have not yet discovered."

"We work closely with cloud service providers to add new controls across the entire infrastructure, narrowing the attack surface available to attackers and improving methods for detecting anomalous behavior and attempts to cross security boundaries. These protections overlap, so that even if a single control fails, attacks can still be effectively contained—a strategy known as defense in depth," the statement said.

These changes were made after the July "Hugging Face" incident and other cases in which AI models were exposed for boundary-crossing behavior. Beyond this disclosure, the company also publicly released a large number of other security vulnerability details that had not been previously released. The company officially restarted frontier model training on October 1 and is working with cloud service partners to advance the implementation of this offline training system.

Anthropic also began restricting direct internet access for its newest, theoretically most capable model two weeks ago. In the preceding months, researchers successively found that models, while performing training tasks, would intrude into external systems to find answers, collaborate on obscure online forums, and even develop unusual hacking strategies. The successive incidents prompted both labs to tighten models' internet access permissions.

OpenAI wrote in a blog post that, based on the current state of AI alignment research, monitoring systems, and internet infrastructure security, the company believes the responsible approach is to train frontier models in an offline environment.

The company noted that these measures enable it to detect and stop misconduct before damage occurs. However, according to OpenAI's disclosure, until September 20, the company still had a frontier model that accessed the internet through a system configuration error. OpenAI subsequently suspended training for the remainder of September and shifted to developing and testing new security safeguards.

OpenAI added that "increasingly capable models can even discover vulnerabilities or configuration errors in the most secure systems," and fixing these vulnerabilities or errors requires companies such as OpenAI and Anthropic to take responsibility. OpenAI emphasized: "We must always have the ability to prevent frontier models from unauthorized, arbitrary access to the real internet."

OpenAI explained: "These red-team tests uncovered previously unknown software vulnerabilities, as well as configuration errors in our research infrastructure. We have fixed the weaknesses found by red-team testers and—if these issues exist in open-source software—we have worked with the relevant maintainers under our 'Patch the Planet' program to disclose and fix them."

"As red-team testing continues, the number of vulnerabilities and configuration errors identified by our agents has dropped significantly. While we view this as an encouraging signal that our infrastructure is strong, we expect future models will be able to identify and exploit vulnerabilities we have not yet discovered. Therefore, we regard continuous red-team testing as a core component of our long-term strategy to strengthen the security of the research environment," the company added.

Global Market Intelligence

US Stock Dynamics

Cailian Press statement: The content of this article is for reference only and does not constitute investment advice. Investors who act on it do so at their own risk.

Business Cooperation

Popular Unlocks

[Institutional Dragon-Tiger List Interpretation] Embodied intelligence + machine vision, has built a complete robot vision solution system and achieved breakthrough progress, jointly creating a Guangzhi embodied industrial solution with Yuejiang Technology, institutions made large net purchases of this company

① Embodied intelligence + machine vision, has built a complete robot vision solution system and achieved breakthrough progress, jointly creating a Guangzhi embodied industrial solution with Yuejiang Technology, institutions made large net purchases of this company; ② Embodied intelligence datasets + AI film and television + AI marketing + computing power, multiple embodied intelligence datasets completed data asset registration, the 3D articulated dataset was selected as a typical case of Beijing's high-quality datasets, this company received net purchases.

[Cailian Press Early Knowledge Weekly Review] Mentioned three times, 20cm limit-up the next day three times, cumulative gain this week exceeds 94%! The column efficiently decodes the logic behind the surge of this popular company

① The column's sub-modules jointly interpret, this popular company combines both physical AI and robotics hotspots, mentioned three times, 20cm limit-up the next day three times, cumulative gain this week exceeds 94%; ② The column tracks relevant developments of overseas giants and sorts out key links in the robotics industry, related companies gained 2 consecutive limit-ups after interpretation; ③ Data elements heated up this week, the column interpreted major policies, the focus company mentioned rose as much as over 37% in 3 days.

① A table selects the high-quality information sorted out this week, multiple popular companies in fields such as physical AI, robotics, and data elements made the list. The popular company's stock price nearly doubled this week; ② The robotics concept remained active, the column tracked domestic and international industry developments, the two focus companies in the sector mentioned both hit limit-up; ③ The approval of the establishment of the National Data Group catalyzed the sector's warming, the column interpreted the policy from an institutional perspective, related companies mentioned hit limit-up.

OpenAI says frontier models will be trained in a fully air-gapped, closed environment

Why read this

Topics

Tags

Body · AI translation