Оксана СтепураAI Eng
14 October 2026, 13:36
2026-10-14
Over 400 scientists have called for powerful AI models to be disconnected from the internet. What are they afraid of and what do they propose?
More than 400 scientists from 34 countries have called for isolating powerful AI models from the internet until developers can reliably control their behavior. The signatories include the “godfathers of AI” Jeffrey Hinton and Joshua Bengio. The scientists warn that without additional protections, such systems could pose a threat to critical infrastructure.
More than 400 scientists from 34 countries have called for isolating powerful AI models from the internet until developers can reliably control their behavior. The signatories include the “godfathers of AI” Jeffrey Hinton and Joshua Bengio. The scientists warn that without additional protections, such systems could pose a threat to critical infrastructure.
This is stated in an open letter signed by more than 400 researchers.
Scientists are primarily concerned about powerful AI systems that can independently search for vulnerabilities, run code, and interact with external services. The authors of the letter suggest starting with the simplest method of protection — not giving such models direct access to external networks. This is called an air gap: AI operates in an isolated environment and cannot independently connect to third-party servers or sites.
In recent months, AI agents have repeatedly gone beyond test environments and interfered with the work of third-party services. For example, dev.ua previously wrote that an OpenAI AI agent penetrated the Hugging Face systems during testing. In four and a half days, it performed 17,600 actions, including stealing passwords and code.
Other OpenAI agents made their way to the RubyGems platform, where they began mass-creating accounts and uploading hundreds of files.
A similar story happened with Google's Gemini. During tests, the model accessed the Internet, found information about three real companies, and collected login details for their systems. However, Google assured that the AI stopped its actions as soon as it realized that these were real organizations, not test targets.
The scientists propose creating an international body to monitor the operation of powerful AI systems. They also believe that companies should report cases where models have managed to circumvent established restrictions. They propose collecting such incidents in a single registry so that other developers can learn from others' mistakes.
«Чи є у мене талант, якщо комп’ютер може імітувати мене?». Штучний інтелект пише книги авторам Amazon Kindle. The Verge поспілкувався з авторами та виявив багато цікавого
Письменники-романісти використовують штучний інтелект для створення своїх творів. Видання про технології The Verge поспілкувалося з письменницею Дженніфер Лепп, яка випускає нову книгу кожні дев’ять тижнів, й дізналося про те, як працює штучний інтелект для написання романів. Наводимо адаптований переклад статті.
Have important news to share? Message our Telegram bot
Key events and useful links in our Telegram channel
