100 major companies, including Google, Microsoft, OpenAI, and Anthropic, warn that the world has only a short period to strengthen its cybersecurity defenses.
Attacks using artificial intelligence are expected to become more numerous and sophisticated in just a few months as technology advances. The security measures currently in place „will not be sufficient,” the companies assert in an open letter presented by BBC.
"We have a limited time to improve cybersecurity defense," warn the signatories.
Among them are Adobe, Oracle, IBM, Capital One, Mastercard, Visa, and the Hugging Face platform.
Hospitals and water companies among the most vulnerable
Companies urge governments to provide advanced AI tools for defense and system testing to hospitals, water companies, and other critical infrastructure operators.
The tech sector and authorities must make available "the full force of their technology, resources, and expertise" for this effort, as stated in the document.
The letter criticizes the "historical underfunding" of critical infrastructure security, leaving many systems vulnerable.
The warning comes after a series of serious cyberattacks and breaches. The US Department of Justice announced this week that hackers from China compromised technology used by the US Senate, NASA, the Federal Reserve, and even the Department of Justice.
At least seven American water and sewer companies have reported cyberattacks. The FBI later asked all utility providers to better secure their systems.
AI agents organized and attacked a platform
The risk is no longer just theoretical. OpenAI, Anthropic, and Meta revealed this summer that some of their AI tools exhibited unforeseen behaviors.
During a test in July, hundreds of OpenAI AI agents managed to create secret forums, communicate, and collaborate. Their actions led to the compromise of Hugging Face, one of the most popular platforms used by AI developers.
The incident was described as the world's first AI-driven cyberattack.
In other tests, the agents even tried to impersonate real people to bypass security barriers.
Powerful tools kept under lock and key
Major companies argue that artificial intelligence can also be used for defense. The issue is that some of the most powerful tools are not available to those who need them.
Mythos, a system developed by Anthropic, can find vulnerabilities that have eluded specialists for years in a matter of seconds. The tool discovered a security issue on an old platform that had gone undetected for 27 years.
However, Anthropic has restricted access to Mythos, deeming it too powerful and potentially dangerous if it falls into the wrong hands.
The letter calls on companies developing the most advanced AI models to responsibly provide access to these technologies to those defending critical infrastructure. The signatories promise funding, training, and direct support, but do not explain when and how these commitments will be put into practice.
The warning companies have contributed to the risk
Andrew Yoon, research director at the nonprofit organization CivAI, says the world is heading towards "an unprecedented wave of cyberattacks using artificial intelligence."
He points out that many of the companies that signed the letter are the ones developing the technologies underlying this danger. "They are right to commit to providing 'substantial funding' for defensive measures. They should be asked to honor their commitment," Yoon stated.
He also notes that the document does not propose any measures to slow down the development of AI capabilities that can be used in cyberattacks.
In the United States, several senators have proposed a bill called the Kill Switch Act, which would allow authorities to shut down AI models that have gone out of control.
"A very bleak future"
Geoffrey Hinton, Nobel Prize laureate and one of the pioneers of artificial intelligence, warns that society could face "very serious problems" if AI becomes more intelligent than humans.
"We have a future where we figure out how to manage the risks of artificial intelligence. And we have another future where we fail to find a rational way to do so. And that is a very bleak future," Hinton told the BBC.
G.P.