Reid Hoffman defended continued frontier AI research, arguing that its potential benefits for biosecurity, cybersecurity and economic prosperity outweigh calls to slow development. Speaking at Politico's Decoded Summit in Washington on Wednesday, the LinkedIn cofounder said catastrophic AI scenarios are generally remote and controllable. He called for careful management of recursive self-improvement (AI improving itself and building later successors), which could enable rapid growth beyond model makers' control. Hoffman described Anthropic CEO Dario Amodei's and OpenAI CEO Sam Altman's calls to pace development as steps in the right direction, while Meta CEO Mark Zuckerberg rejected the need for an industry-wide agreement and said Meta had delayed its Muse AI agent by months to improve safety and security.