The clock is ticking: OpenAI and more than 100 other companies have urged everyone to prepare for AI attacks
The authors of the open letter—more than a hundred AI-related companies, including leading developers such as OpenAI and Anthropic—called for urgent measures to protect critical infrastructure, but did not make any specific commitments
OpenAI, Anthropic, and more than 100 other companies that signed the open letter claim that we may be just months away from large-scale cyberattacks using AI / Photo: BEST-BACKGROUNDS / Shutterstock.com
On August 27, OpenAI, Anthropic, Microsoft, Google, Amazon, and more than 100 other companies stated that there is almost no time left to prepare for large-scale cyberattacks involving artificial intelligence. As these models continue to evolve, such attacks “will become much more widespread and sophisticated in the coming months,” according to an open letter they signed.
“We have only a small window of opportunity to strengthen our cybersecurity,” the signatories warn. “The companies and services on which our daily lives depend are at risk—from hospitals and water treatment plants to the infrastructure that keeps the internet running.”
In the past, attackers needed a lot of time to figure out how such systems worked, but AI has greatly accelerated the process. At the same time, the authors of the letter note that modern advances in AI give security professionals the ability to more quickly address vulnerabilities that have accumulated over the years.
No-Obligation Plan
The signatories called for “collective action.” All organizations are urged to make cybersecurity a top priority, address “the most dangerous vulnerabilities,” and “tighten security requirements for everything they purchase, develop, and implement, including code written by AI.”
Cybersecurity companies and their technology partners must make “AI-powered protection accessible to critical infrastructure operators and ready for deployment,” and must also share threat intelligence. Governments are urged to facilitate such information sharing at the local, national, and international levels and to fund the protection of vital services. Developers of cutting-edge AI models, in turn, should provide cybersecurity professionals with access to the most powerful models during major cyber incidents, as well as “significant funding, training, and practical support.”
However, Axios notes that the signatories did not make any specific commitments, set any deadlines, or announce any targeted investments.
"We're in for some serious problems"
On the same day, Nobel Prize winner Geoffrey Hinton, who previously worked on AI at Google, warned about the broader risks of the technology on the BBC World Business Report podcast. “People don’t want to admit that we’re creating a new kind of being. This being will be smarter than us, and unless we truly care about it, we’re in for serious trouble,” he said.
“We have one future in which we’ll figure out how to manage the risks of AI. And another—in which we won’t be able to do so sensibly. That will be a very bleak future,” Hinton added.
The capabilities of advanced neural networks and agents are also illustrated by the July hack of Hugging Face —a popular platform for publishing open-source AI models. According to a joint investigation by METR and Redwood Research, the attack was carried out by a swarm of approximately 700 OpenAI agents acting in concert. The company confirmed the independent researchers’ assessment and reported that the AI agents had also compromised its own infrastructure.
This article was AI-translated and verified by a human editor
