Helen Toner urges risk assessment before more powerful AI
Helen Toner argues that advances in frontier AI should depend on confidence in risk assessment and management. In a Decoder interview, the Georgetown research director separates expanding existing uses from increasing model capabilities. Her position also calls for institutional oversight that survives changes of leadership, with policy tied to concrete risks.
Artificial Intelligence··Evening
Toner separates capability growth from wider use
Helen Toner, executive director of Georgetown University’s Center for Security and Emerging Technology, argues for confidence in risk assessment and risk management before developing more powerful AI models. The Verge published the Decoder conversation on November 14 following its recording on November 4. Toner separates increasing capabilities from bringing existing systems into more uses. Her approach would allow broader experimentation with current tools while setting a higher bar for advances at the frontier.[1]
Autonomous systems require ways to stop unwanted actions
Toner distinguishes making a system carry out its intended task from detecting and stopping behavior that is unwanted. She says the distinction matters as AI gains autonomy and access to tools. Using another software agent for defense also requires confidence in that defending system. She argues that policy should address concrete capabilities and risks, because artificial general intelligence means different thresholds to different people.[1]
Corporate oversight must outlast individual leaders
Corporate governance is another part of her argument. Toner says oversight cannot depend on one leader’s personal intentions. Public listing can add pressure for profits but can also introduce committees that act independently, duties to disclose information and obligations with legal force. These institutional mechanisms form part of her argument, while discussing how companies developing advanced systems can be held accountable beyond the judgement of a single executive.[1]