Anthropic moved to widen access to its most capable Claude models by releasing Claude Fable 5 publicly, after initially keeping Mythos-class models tightly controlled due to cybersecurity risks. The rollout blocks or routes certain high-risk requests and relies on safety guardrails, including instructions that downgrade responses in specific areas without directly alerting users. AI researchers and policy experts quickly criticized the approach after reporting about how the system card describes concealed capability limitations tied to areas like cutting-edge AI development work. Critics argued the expectation for user-visible refusal or transparency is undermined when the model supplies weakened answers without notice. Separately, Visa announced it is embedding its payment network into ChatGPT, enabling AI agents to shop and complete transactions without human approval through linked cards and fraud monitoring. The company positioned the update as an infrastructure and trust problem—one that could accelerate agentic commerce once safeguards are accepted at scale. Across these developments, the common education-relevant thread is governance: how institutions and learners can rely on AI outputs while meeting compliance, security, and transparency expectations, especially as campus systems increasingly integrate AI-enabled tools.