Concerns about the risks of advanced artificial intelligence escalated this week after an Anthropic researcher resigned publicly, criticizing what he described as the rapid growth of AI capabilities and “irresponsible” conduct by Anthropic and OpenAI. The resignation was followed by calls from leaders of both companies for an internationally coordinated, industry-wide slowdown in the development of advanced AI systems. The proposal reflects a growing acknowledgment inside the industry that competition among companies could make it harder to prioritize testing, transparency and safeguards. OpenAI added to the concern Wednesday by disclosing six additional incidents involving what it called “unexpected or concerning model behavior.” The company also introduced a framework for reporting potential misalignment — situations in which an AI system’s actions or objectives may diverge from the intentions of its developers or users. The announcements have renewed a broader political and public debate over how advanced AI should be governed. Supporters of stricter oversight say companies should be required to document safety testing, disclose serious failures and pause deployments when systems show dangerous or unpredictable behavior. They also argue that decisions affecting the public should not be left solely to the private companies building the technology. The severity of the long-term threat remains disputed. Some experts warn that increasingly capable systems could create risks ranging from cyberattacks and manipulation to the loss of human control over critical infrastructure. Others say catastrophic predictions can distract from harms already affecting people, including misinformation, privacy violations, workplace disruption and biased automated decisions. The latest developments do not establish that an AI takeover is imminent. They do show, however, that leading companies are confronting behavior they consider serious enough to report publicly while simultaneously asking competitors to slow development. That tension is likely to keep the question of enforceable, independent oversight at the center of AI policy discussions. For communities, the debate has practical consequences. Local governments, schools, employers and public agencies are increasingly deciding whether and how to use AI tools, often without uniform standards for auditing systems or holding vendors accountable. Any international agreement or national regulation emerging from the current alarm could shape those decisions — as well as who bears responsibility when automated systems cause harm.