Yesterday saw multiple new developments in the Anthropic Mythos\/Fable 5 export control incident, with dense information flow. WIRED reported that Fable 5 will not return\u2014Trump administration officials stated that if Anthropic wants to re-release, it must ensure model guardrails cannot be bypassed, which security experts deem impossible. The New York Times obtained and reported internal Anthropic chat messages, where employees felt “unfairly targeted” and “like they don’t want us to exist.” White House AI head David Sacks acknowledged that open-source models with Mythos-level capabilities will eventually spread widely, while also stating the White House overstepped in banning all foreign national researchers at Anthropic from using Fable 5. Additionally, @MaxForAI cited WIRED’s report revealing the SK Telecom timeline: days before Amazon’s complaint, the White House had called Anthropic to revoke SK Telecom’s access to Mythos, citing suspected ties to China. At the G7 AI lunch, Anthropic founder Dario Amodei and Google DeepMind founder Demis Hassabis called for a US-led AI alliance excluding countries like China.<\/p>\n
Sources:<\/p>\n
Noam Shazeer, core author of the Transformer paper and co-founder of Google Gemini, officially announced his move to OpenAI, marking the most notable AI talent shift this week. Shazeer previously founded Character.AI, returning to Google via acquisition as Gemini’s technical lead. @MaxForAI noted the impact on Gemini’s product line (especially the unreleased Gemini 3.5 Pro) remains unclear.<\/p>\n
Sources:<\/p>\n
DeepSeek founder @PKUCXK announced that Vision capabilities are now available on both web and app, with @MaxForAI dubbing it DeepSeek V4.1. This marks DeepSeek’s first release of image recognition to general users, continuing its multimodal expansion beyond text-only models.<\/p>\n
Sources:<\/p>\n
On June 18, Cailianshe reported that eight ministries, including the Ministry of Commerce, jointly issued implementation opinions on accelerating “AI + Consumption” development. The document proposes using personal consumer loan fiscal discount policies to support consumers purchasing AI-related products, encouraging financial institutions to develop products and services around consumer demand for AI purchases. Multiple bloggers noted this is the first time the government has supported AI product consumption through personal consumer loan channels.<\/p>\n
Sources:<\/p>\n
xAI’s ecosystem saw multiple updates yesterday. xAI released Grok Imagine Video 1.5, an image-to-video model boasting clearer realistic effects, better physics simulation, and faster generation. Grok Build released v0.2.55 with several optimizations. Additionally, Grok is now available on major cloud platforms including Bedrock, Vertex, Oracle, and Azure. @elonmusk stated that through real-time access to X data, Grok is rapidly improving. User @milichab demonstrated Grok Build 0.1’s code review capabilities, with Musk replying “Progress.”<\/p>\n
Sources:<\/p>\n
Multiple bloggers noted that ByteDance’s video generation model Seedance 2.5 is set to release in early July. Currently, no text-to-video model approaches Seedance in quality. This announcement has sparked discussion in the Chinese AI community.<\/p>\n
Sources:<\/p>\n
@MaxForAI noted that Huawei has successfully registered a related trademark, and reminded domestic AI application developers to avoid using phrases like “world’s first” to prevent trademark disputes. The topic has been widely shared in the Chinese developer community.<\/p>\n
Sources:<\/p>\n