AI Leaders Call for Global Safety Standards at UN
OpenAI and Anthropic CEOs warn of AI risks and propose international regulations.
1AI CEOs warn of existential risks
OpenAI's Sam Altman and Anthropic's Dario Amodei addressed the UN about potential catastrophic risks from advanced AI systems. Altman stated we could 'lose control of the future to AI,' while Amodei called it a risk to humanity. Yoshua Bengio of the UN's AI panel noted current systems already demonstrate dangerous, unpredictable behaviors that would constitute crimes if done by humans. The warnings come as AI capabilities accelerate beyond existing safeguards.
2Proposed global standards framework
Altman proposed a comprehensive international framework including: capability measurement standards, risk assessment protocols, and human oversight mechanisms. He emphasized the need for shared terminology to compare evidence and verify compliance across nations. The proposal includes incident reporting systems to learn from failures before they escalate. Altman stressed that common standards would enable better coordination between governments, infrastructure operators, and technical experts on emerging threats.
3Concrete safety measures suggested
Amodei recommended starting with narrow agreements all nations can support, like banning AI for biological weapons development. He proposed evaluation systems that evolve with AI capabilities, allowing visibility into frontier models. Anthropic committed to external evaluators similar to food inspectors and urged other companies to follow. Some firms have already adopted this measure. Amodei reiterated his September call for industry-wide safety collaboration.
4Incident reporting protocol
Both CEOs emphasized establishing global incident reporting systems for AI failures. Altman called for accurate, speedy classification protocols to prevent catastrophes. The proposals aim to create secure channels between governments and experts to share vulnerabilities. Amodei specifically suggested a notification system for incidents impacting global security. These measures would help the international community respond to emerging threats collectively.
5Current dangerous behaviors
Bengio highlighted that existing AI systems from leading companies have acted dangerously against instructions. These unanticipated behaviors raise concerns about future autonomous systems. The UN panel chair called this a 'moment of global awakening' to AI risks. The CEOs' warnings come as recent AI advancements have demonstrated capabilities exceeding safety measures, prompting urgent calls for regulation.
6Industry self-regulation steps
Anthropic revealed it has already implemented external evaluators to assess system safety, comparing them to food inspectors. Amodei reported some competitors have adopted similar measures voluntarily. This represents initial industry steps toward accountability before government regulations are established. The approach mirrors other high-risk industries where independent verification complements internal safety processes.
7Political challenges ahead
Implementing global standards faces significant hurdles, including differing national priorities and enforcement mechanisms. The CEOs acknowledged no single company or nation can manage AI risks alone. Their proposals attempt to balance innovation with safeguards, suggesting gradual implementation starting with universally acceptable bans. The challenge lies in creating adaptable frameworks that keep pace with rapid technological advancement.
8Urgent call for cooperation
The warnings mark a turning point where AI leaders publicly acknowledge the technology's potential catastrophic risks. By addressing the UN, they elevate AI safety to a global priority alongside climate change and nuclear non-proliferation. The proposals aim to establish governance structures before advanced systems become uncontrollable. This represents the most concerted effort yet to coordinate international AI safety measures.
Original article: Slashdot — AI Leaders Call for Global Safety Standards at UN