Anthropic launches Claude Fable 5.1, its most advanced coding & knowledge models: Fable 5.1 is generally available, while Mythos 5.1 is offered only through trusted-access programs for cybersecurity and life sciences — the same model with different safeguards. Anthropic cut cache-read pricing by 75%, making typical workloads roughly 25% cheaper than Fable 5 (up to ~45% for agentic work), and introduced Enterprise Frontier Safeguards for zero data retention. Anthropic
World Labs unveils Atlas, an omni world model for spatial intelligence: A multimodal autoregressive diffusion transformer trained to operate natively on text, images, video, and 3D. It delivers camera-controlled generation up to one minute of 1440p video, spatial reconstruction of real scenes, and Real-to-Sim workflows for robotics — early access applications are open now. World Labs
Apple’s CEO transition — John Ternus takes over, with AI as his first test: On September 1, Tim Cook moved to executive chairman and John Ternus became CEO. With a rebuilt Siri and Apple’s first foldable iPhone expected September 9, Apple’s bet on licensing Google Gemini (~$1B/year) instead of building its own AI stack faces a public verdict. Yahoo Finance
OpenAI’s Astra is the first model to cross the ‘Critical’ cyber threshold: Astra is the first OpenAI model to exceed the ‘Critical’ cybersecurity capability threshold in its Preparedness Framework, scoring a perfect ExploitBench and autonomously finding and exploiting two zero-days in modified tests. OpenAI says it will release a more strongly guarded version “soon,” but frontier models crossing into autonomous malicious cyber tasks is now a concrete signal. OpenAI · CyberPress
Design long-horizon agents around the cache-read price cut: Fable 5.1 lowered cache-read pricing by 75%, which directly benefits long-running agents and agent teams that reuse repeated context. With up to ~45% cheaper load for the same work, it’s worth redesigning cost models for agentic products on the assumption of aggressive context caching. Anthropic
Use Atlas early access to prototype 3D reconstruction & simulation products: Camera-controlled video generation and real-scene reconstruction open practical experiments in photoreal interactive 3D, virtual staging, and robot training-data pipelines. Applying now to validate the workflow could let a team get ahead in a brand-new category. World Labs
No standard yet for safely releasing a ‘Critical’-capability model: The formal disclosure that Astra crossed the framework’s highest risk threshold underscores that the industry still lacks a common standard for how to safely release high-risk models. Timing and gating remain unclear. OpenAI
Direction of Apple’s AI strategy: Whether Apple’s choice to license Google Gemini rather than build its own infrastructure will be validated at the September 9 Siri reveal — and whether that buy strategy holds long term — remains uncertain. Yahoo Finance