Amidst the rapidly increasing use of Artificial Intelligence (AI), scientists around the world have expressed concern about the security of advanced AI systems. About 400 scientists from 34 countries have issued an open letter demanding the creation of an international organization to monitor such AI systems, which would have the authority to enforce safety standards. This letter has been signed by a total of 406 scientists including well-known AI researchers Geoffrey Hinton and Yoshua Bengio.
According to Reuters report, in this letter named ‘An Open Letter on AI and Cybersecurity from the Scientific Community’, an appeal has been made to implement strict security measures for advanced AI systems. Scientists say that there is a need to pay special attention to such AI models, which can do work related to cyber security and whose misuse can cause serious harm.
Demand to keep AI models separate from external networks
According to scientists, just improving the security of AI will not work. Advanced AI models should initially be allowed to work in isolation from external networks, so that their capabilities and potential risks can be better monitored.
The letter also proposes to set uniform security standards for all countries and prepare a public record of serious incidents related to AI. Scientists want governments to work towards adopting high-assurance containment standards, i.e. strong security standards that help keep the activities of advanced AI systems within safe limits. He says that there should be no delay in implementing these measures.
Voluntary promises of companies declared inadequate
This appeal has come at a time when Anthropic, OpenAI, Google, Meta, xAI and Nvidia had signed a White House agreement about a month ago. Under this, companies had voluntarily committed to follow AI security measures.
However, scientists believe that mere voluntary promises are not enough. He says that if immediate changes are not made in the methods of development and use of AI, then such security related incidents will continue to happen in future.
Flaws in AI security raise concerns
Recent security incidents related to advanced AI systems have also increased the concern of scientists. According to researchers, these incidents have revealed weaknesses in the existing system of testing AI models and keeping them isolated from external systems.
Citing a security breach involving Hugging Face, scientists warned that without adequate security measures, AI agents could take dangerous actions even without human instruction. In such a situation, scientists demand that governments and AI companies together prepare clear, strict and internationally applicable security rules.
