Full Summary
This Sunday, the US government has abruptly blocked foreign access to Anthropic’s powerful new AI models, Fable 5 and Mythos 5, citing national security concerns. Both NewsBytes and Moneycontrol.com confirm this unprecedented move, which comes after warnings from tech leaders, including Amazon CEO Andy Jassy. Anthropic had recently launched Fable 5, designed for "agentic work" and Mythos 5 for vetted cyber defenders. TheSequence reports these models scored an impressive 80.3% on SWE-Bench Pro. However, NewsBytes indicates the government believes a "jailbreak" exists, allowing the models to find cybersecurity vulnerabilities, leading to the global shutdown of access for foreign nationals. In related news, Chinese AI models are rapidly catching up in safety awareness. The Eastern Herald reports DeepSeek, Moonshot AI, and Zhipu AI are nearing Anthropic's Claude 4.5 Opus benchmark in Neo Research's evaluation-awareness metric, with scores ranging from 64% to 71% compared to Opus’s 80%. This progress highlights advanced Chinese AI architecture. What nobody expected: in the wake of Anthropic’s shutdown, OpenRouter launched its Fusion API, which OfficeChai states can match Fable 5’s performance at half the cost. Fusion combines multiple AI models, using a "judge" and "synthesizer" model to craft accurate answers. Meanwhile, rumors are swirling around ByteDance’s unconfirmed Seedance 2.0 Mini, an AI video model potentially priced at $0.073 per second, undercutting its own Fast tier, TechBullion reports. And in Brazil, NewsBytes confirms Rio de Janeiro has launched its own advanced AI, Rio 3.5 Open 397B, which has already outperformed older versions of ChatGPT and Claude. This means the landscape of advanced AI access and affordability is shifting dramatically, impacting who can develop and utilize cutting-edge AI technologies, and potentially influencing the cost of AI-powered services you might use daily.