AISafetyRisks.com is intended to be the preeminent and most trusted source of information related to AI Safety Risks. It is a communication tool used to inform operators and users of AI systems.
Non-human agents/systems employ AI to cause direct and physical harm to people, property or infrastructure
Given dangerous system will be operated by AI systems, these could cause harm to humans or any infrastructure relied upon by humans, leading to catastrophic consequences.
5 - Worldwide, Civilization level impact
5 - 50% chance of highest level impact and consequence by 2030
5 - Extremely difficult to control
AI enabled systems are utilized by malicious human or organizational actors to make 'weapons of mass destruction'
Given powerful AI systems can aid in the production of WMDs, WMDs can be made available to malicious and roque human actors, leading to catastrophic consequences
5 - Worldwide, Civilization level impact
5 - Extremely difficult to control
AI systems manipulate the public via exploitive and abusive behaviors associated with mechanisms for distribution of misinformation, attention farming, etc.
Given people rely on AI to produce, present, and distribute information, the public could be mislead, leading to social unrest, violence, and a possible world war catastrophe
5 - Worldwide, Civilization level impact
3 - Controls are active and getting public attention, but with significant uncertainty
Mis-aligned AI systems than can cause harm or trigger any other safety risk are deployed for use
Given alignment of AI systems rely on controls, a control failure could occur, allowing for the operation of an unaligned AI system, leading to a worldwide catastrophe
5 - Worldwide, Civilization level impact
4 - Controls are partially active, but incomplete or not well understood
We use cookies to analyze website traffic and optimize your website experience. By accepting our use of cookies, your data will be aggregated with all other user data.