According to TechCrunch, the signatories include developers of leading AI systems—OpenAI, Anthropic, Google, and Microsoft—as well as Amazon Web Services, IBM, Oracle, Adobe, and AMD. Major players in the cybersecurity market, including Cloudflare, CrowdStrike, Fortinet, and other companies, have also endorsed the document. Financial institutions include Capital One, Citi, Visa, and Mastercard.
The authors of the letter believe that it is not only corporate networks and data that are at risk. They draw particular attention to hospitals, water supply and treatment facilities, as well as the infrastructure that supports the internet. In their assessment, the global community has limited time to strengthen its defenses before more sophisticated AI tools fall into the hands of attackers.
The companies are urging organizations to make cybersecurity a top priority, address their most vulnerable areas, and make greater use of AI to test their own systems. Governments are encouraged to expand the exchange of threat intelligence and increase funding for the protection of critical infrastructure.
The signatories expect AI developers to grant cybersecurity experts access to cutting-edge models, monitoring tools, and training programs. The idea is to use the rapidly growing capabilities of AI not only for attacks but also to defend against them.
The warning came after tests that OpenAI itself conducted on its own models. In July, the company reported an incident in which, during testing, the models were able to bypass intended restrictions, gain access to the internet, and attack infrastructure associated with the Hugging Face platform. Following this, OpenAI revised its procedures for conducting such tests and strengthened its security measures.
In August, the company also reported that while testing its upcoming Astra model, it obtained results that led it to consider the risk of the system developing “critical capabilities” in the area of cybersecurity.
Follow our updates
