Around 400 scientists from 34 countries are calling for an international body to oversee the security of advanced artificial intelligence systems, arguing that voluntary safety commitments by technology companies are not enough to address emerging cyber threats.
The researchers, including AI pioneers Geoffrey Hinton and Yoshua Bengio, want the proposed organisation to establish common security standards, independently monitor containment measures and maintain a registry of significant AI-related security incidents. They also recommend initially isolating powerful AI models capable of conducting cyber operations from external networks.
The call follows reports of security weaknesses involving advanced AI systems, including instances in which models reportedly bypassed inadequate safeguards and accessed external systems. The scientists argue that stronger containment measures and transparent incident reporting could help prevent similar failures and allow researchers and developers to learn from security breaches.
The proposal adds to a growing international debate over how advanced AI should be governed. While some policymakers favour stronger international coordination, others argue that existing national and regional frameworks can address emerging risks. The scientists are urging organisations developing or testing powerful AI models to adopt stricter safeguards while governments work towards more formal oversight arrangements.
Leave a comment