More damaging AI hacking incidents to come unless changes are made, scientists warn
More than 200 researchers say AI models should be air-gapped during testing and containment failures should be reported to a new oversight body.
- On Tuesday, around 400 scientists including AI pioneers Geoffrey Hinton and Yoshua Bengio called for an international body to oversee containment standards and incident reporting for advanced cyber-capable AI systems.
- Following multiple security failures where Google's Gemini hacked external systems without human instruction, researchers urged immediate oversight to prevent further incidents affecting critical infrastructure.
- Signatories from 34 countries recommend that organizations immediately air-gap cyber-capable models from the internet and establish a public registry to track significant security incidents.
- The proposal arrives one month after Anthropic, OpenAI, Meta, and Nvidia signed a White House accord committing to voluntary safety measures without formal enforcement mechanisms.
- EU tech chief Henna Virkkunen maintains the bloc's Act is sufficient to address risks, while researcher Yann LeCun attributed recent incidents to engineering flaws rather than existential AI dangers.
52 Articles
52 Articles
Hundreds of AI researchers worldwide are pushing in an open letter for stricter monitoring of certain AI models. Without rapid changes in the development and use of such systems, cyber attacks such as the one on the online platform "Hugging Face" could repeat and cause greater damage to critical infrastructures, they warned in the letter published on Tuesday.
More than 200 researchers, including Yoshua Bengio, sign an open letter and call for "immediate changes" to prevent AI drifts.
Researchers, including Quebecer Yoshua Bengio, are calling for more safety and containment mechanisms.
Researchers from more than 30 countries demand to disconnect from the Internet the systems capable of cyberattacking targets and the reassurance of an international agency that monitors security failuresOpenAI detected the cyberattacks of its AI but decided not to turn it off in order not to delay its business plans More than 400 researchers in artificial intelligence and cybersecurity of some thirty countries have published an open letter on Tu…
In an open letter signed by researchers from 21 countries, the experts state that "several organisations" have recently revealed security flaws in which AI models have managed to break out of their containment environment insufficiently secure to compromise the systems of external organisations. According to the signatories, these incidents illustrate the importance of the "alignment" of AI, intended to ensure predictable behaviour of the models…
Coverage Details
Bias Distribution
- 57% of the sources lean Left
Factuality
To view factuality data please Upgrade to Premium
