OpenAI, Anthropic, Google, Microsoft, and over 100 other companies and organizations are warning that AI-powered cyberattacks could become much more widespread and sophisticated in the coming months. They say that even hospitals, water utilities, and other critical infrastructure could be under attack.
OpenAI, Anthropic, Google, Microsoft, and over 100 other companies and organizations are warning that AI-powered cyberattacks could become much more widespread and sophisticated in the coming months. They say that even hospitals, water utilities, and other critical infrastructure could be under attack.
This is stated in a joint open letter published on the OpenAI website.
The signatories include OpenAI, Anthropic, Google, Microsoft, AWS, Cloudflare, CrowdStrike, Fortinet, Okta, Palo Alto Networks, Mastercard, Visa, IBM, Oracle, Figma, Shopify, and others. Among them is Hugging Face, a company whose systems were recently attacked by OpenAI’s AI agent during an internal experiment. The authors of the letter believe that the old approach to cybersecurity, where companies gradually find and close vulnerabilities, may simply not keep up with the new capabilities of AI.
The authors of the letter believe that the old approach to cybersecurity — where companies gradually find and close vulnerabilities — may simply not keep up with the new capabilities of AI. After all, corporate systems accumulate bugs, weak passwords, excessive access rights, incorrect settings and outdated software over the years. However, in the past, an attacker had to spend a lot of time finding such weaknesses. Now, some of this work can be automated by AI systems.
Organizations are advised to close critical vulnerabilities more quickly and to more closely verify code, including that created by AI. Cybersecurity companies should also test their products with the capabilities of the strongest models and more actively share information about attacks.
In return, AI developers are expected to give proven cybersecurity experts access to their most powerful models, as well as provide them with funding and training. Governments are urged to invest more in protecting critical infrastructure and to share information about attacks between countries.
Some of the signatories are now developing their own models that work with cybersecurity. OpenAI this month expanded its Daybreak program and introduced GPT-5.6-Cyber for authorized cybersecurity tasks. Anthropic is developing the Claude Mythos model and the Glasswing project to find vulnerabilities in critical software, and Microsoft has launched Project Perception, a system of AI agents that look for vulnerabilities, analyze attacks, and help remediate them.
Previously, dev.ua wrote that the head of Hugging Face asked OpenAI for $100 million to strengthen cybersecurity after the company's autonomous AI agent gained access to the platform's internal data.
