487 AI researchers demand air-gapped training for cyber-capable models
Hundreds of experts warn current security is inadequate, urging new international oversight and strict containment for AI development.
Key points
- AI specialists signed an open letter warning that current training security procedures are inadequate.
- Signatories including Geoffrey Hinton and Yoshua Bengio call for air-gapped systems to contain cyber-capable AI models.
- Researchers propose a new international body to set safety standards and report containment failures.
Hundreds of artificial intelligence researchers have issued a formal warning to governments and the technology community, stating that current security procedures for training AI models are inadequate. In an open letter published recently, 487 specialists, including prominent academics, called for stricter regulations and the creation of a new international oversight body to govern AI development best practices.
The signatories argue that strong sandboxing mechanisms, specifically air-gapping or disconnecting systems from the internet, are necessary to provide high-assurance containment for cyber-capable AI models. This recommendation aims to prevent these powerful tools from being used in malicious cyberattacks or causing unintended security breaches during their development phase.
Calls for International Oversight
The researchers propose establishing a new international body to oversee AI safety standards. This organization would be set up in consultation with governments and a wide range of AI and security specialists. Its primary functions would include defining best practices and serving as a central point for reporting containment failures.
According to the letter, this reporting mechanism would improve transparency for learning and accountability purposes. By centralizing failure data, the industry could better understand risks and refine containment technologies. The experts emphasize that this structure is essential for managing the unique security challenges posed by advanced AI systems.
Context and Industry Reaction
These demands go beyond recent calls for general AI regulation. Earlier this week, Microsoft president Brad Smith noted that AI is less regulated than washing machines or milk, though he offered no specific proposals to fix the gap. The UK parliament is also examining ways to improve AI security, but progress may be slowed by disagreements among AI businesses.
Geoffrey Hinton and Yoshua Bengio, two of the letter’s signatories, previously signed an open letter in October last year warning about the dangers of super-intelligent AI. That earlier initiative did not appear to slow the rapid development of AI technologies. It remains unclear whether corporate leaders will heed this latest warning regarding containment and safety.
What should you do?
Security professionals should monitor developments in AI safety standards and containment protocols. Organizations developing or using AI models should review their current isolation strategies and consider how air-gapping might apply to their specific use cases. Stakeholders should also watch for the formation of new international bodies that could define future compliance requirements for AI safety and containment failures.
What to do and how to stay safe: AI safety
- Review current AI model training environments to ensure they are isolated from production networks and the public internet.
- Monitor official announcements from governments and international bodies regarding new AI safety standards and containment regulations.
- Assess your organization's risk exposure to cyber-capable AI tools and document any potential containment failures for internal review.
- Engage with industry groups to share best practices for sandboxing mechanisms and high-assurance containment technologies.
Step-by-step guide: Social Engineering Warning Signs and How to Respond
General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.
Frequently asked questions
Who signed the open letter regarding AI security?
487 AI specialists, including senior academics Geoffrey Hinton and Yoshua Bengio, signed the letter warning governments and the AI community.
What specific containment measure do the researchers recommend?
The signatories recommend using strong sandboxing mechanisms, including air-gapping or disconnecting systems from the internet, for cyber-capable AI models.
What is the proposed role of the new international body?
The proposed body would set new safety standards and act as an organization for reporting containment failures to improve transparency and accountability.
