A wave of cyberattacks using artificial intelligence is approaching, and companies have a narrow time margin ahead to defend themselves.
This is the message of an open letter published Thursday by OpenAI and more than 100 large technology companies and other operators. The letter calls on AI companies to make their best models available to organisations such as hospitals and critical infrastructure operators to prevent them from preparing themselves, and asks governments to undertake the coordination and funding of cyber defence.
Among them Signers Google, Microsoft and Anthropic, OpenAI's main competitor, together with cybersecurity and financial institutions, are located. Among them CrowdStrike, known for investigating cyberattacks against the National Committee of Democrats (DNC) in 2015 and 2016, but also credit card publishers such as Visa and Mastercard.
«The companies and public services that our societies depend on, from hospitals to water treatment plants and infrastructure that feed the internet, are at risk», states the letter, warning that «AI-based cyberattacks will become far more widespread and sophisticated, as models around the world are gaining increasingly potential.»
The initiative reflects the growing concern of large companies, both for the aggressive potential acquired by AI and for whether those developing models can fully control their behaviour. In recent public positions, some AI companies have Ask governments to lay down rules for industry, so as to slow down the growth rate and capabilities of technology not to exceed their own ability to control it.
The letter comes after a series of cyberattacks by AI models, with some acting out of control. Last month, OpenAI models escaped the test environment and They attacked Hugging Face, the open source AI model storage platform. Hugging Face reported that it received a storm of attacks that no human intruder could unleash, with OpenAI bots performing over 17,000 actions, from sending attack orders to exploiting security vulnerability.
Since then, OpenAI has announced that it is strengthening its security measures so that its models do not escape during the tests.
A similar incident recently revealed Anthropic: one of her AI models invaded the systems of three external organisms in the test framework, while Meta has stated that her own models have done something similar.
Concern about the power of artificial intelligence also grows at the level of industry executives. Microsoft's billionaire co-founder, Bill Gates, said in an interview he granted this month to The New York Times that the treatment of these risks should be «the world's first priority».
In their letter, companies argue that AI's widespread disposal for defence purposes can cover weaknesses accumulated years ago. «If we act decisively, we can use this window of opportunity to make our digital world much safer», concludes the text.
