1. Anthropic cuts off Fable 5 and Mythos 5 access following government order
요약: The Verge reported on June 13 that Anthropic cut off access to Fable 5 and Mythos 5 after a U.S. export-control directive citing national security concerns. The order affected foreign nationals and forced Anthropic to disable the models broadly, including for customers. Anthropic said it was complying but argued the government had not provided specific details, and that reported jailbreak concerns appeared minor and not unique to Mythos. The move matters because it turns frontier-model safety and export control from a policy debate into an immediate operational risk for AI labs, enterprise customers, and globally distributed research teams.
2. OpenAI says it is committed to learning as states investigate ChatGPT's impact on young users
요약: Business Insider reported on June 13 that a coalition of U.S. state attorneys general is investigating OpenAI over ChatGPT's impact on minors and vulnerable users. New York Attorney General Letitia James reportedly subpoenaed documents covering engagement, retention, data handling, health and consumer data, and activity involving young and older users. OpenAI said it takes the concerns seriously, will engage constructively, and has added more protective experiences for minors and people in distress. The investigation raises the compliance bar for consumer AI products, especially around safety design, data governance, and mental-health-adjacent interactions.
3. The AI price war is here, piling pressure on OpenAI and Anthropic
요약: The Wall Street Journal reported in the past week that enterprises and startups are increasingly mixing cheaper open-source and Chinese models into production workflows, reserving premium systems from OpenAI and Anthropic for the hardest tasks. The shift can sharply reduce AI-assisted work costs and puts pressure on frontier labs whose economics depend on high usage and premium pricing. The strategic takeaway is that model selection is moving from leaderboard performance alone toward cost per completed task, latency, reliability, and orchestration across multiple providers.
Sources: