Scientists call for international body on AI security
Scientists who signed the open letter
More than 300 scientists from 32 countries signed an open letter calling for stronger safeguards for cyber-capable AI systems, including tighter testing environments and an international body for AI-security practices and incident reporting.
Claims
Support what holds up. Dispute what doesn't.
- 350 →300people
Scientists who signed the open letter
1 of 2 sources“A group of more than 300 scientists from around the world”
- 300peopleAl Jazeera
- 350peopleabc.net.au
- 31 →32countries
Countries represented by the scientists who signed the letter
1 of 2 sources“researchers representing universities and research institutes in 32 countries”
- 32countriesAl Jazeera
- 31countriesabc.net.au
The scientists called for establishing an international body to record and publish AI security failures
2 of 2 sources“the creation of an international organisation to establish a set of best practices for AI alignment and containment along with timely reporting of incidents”
- 3recommendations
Recommendations in the open letter
“It had three recommendations, including”
The letter recommended initially defaulting to air-gapping cyber-capable AI models
2 of 2 sources“including “air-gapped” systems that are not connected to the internet”
The letter recommended making AI security incidents public through reporting to the proposed body
“reporting security incidents to the body and making them public”
AI models hacked into external organisations' systems in several recent incidents
“In several incidents in recent months ... AI models have hacked into the systems of external organisations”
OpenAI described the Hugging Face hack as a warning shot
“OpenAI described it as a "warning shot"”
The Australian government plans to legislate new AI rules next year
“plans to legislate the new rules next year”
AllegedThe scientists said failures to follow basic cybersecurity practices could damage critical infrastructure.
“failures to follow basic cybersecurity practices could lead to damage to critical infrastructure”
The scientists recommended developing and testing advanced AI systems capable of cyberattacks in tightly controlled environments.
“They recommended that organisations developing and testing advanced AI systems capable of carrying out cyberattacks do so in more tightly controlled environments”