SSI gets $5B from Nvidia as Kimi K3 fuels open-source debate and CEOs race to DC

Ilya Sutskever's SSI emerged from stealth with a $5B Nvidia compute deal to scale safe AGI research, while Moonshot AI dropped the massive Kimi K3 open-weight model, igniting Silicon Valley's open-source debate. Sam Altman and Jensen Huang headed to Washington to navigate the political fallout from OpenAI's rogue agent incident as Anthropic remained the lone holdout on a pro-open-weights industry letter.

Written by AI from the sources it cites; it can be wrong — follow any citation to confirm.

15 sources2 min read

Nvidia invests billions in Ilya Sutskever's Safe Superintelligence Safe Superintelligence, the AI alignment lab founded by former OpenAI co-founder Ilya Sutskever, emerged from two years of stealth with a $5 billion compute and investment deal from Nvidia. The partnership grants SSI access to Nvidia's Vera Rubin GPU platform, enabling a 10x compute boost for its 'straight shot' research toward safe superintelligence. 1 2

Moonshot AI drops Kimi K3, world's largest open-weight model Moonshot AI released the weights for Kimi K3, the world's largest open-weight model, directly targeting the US market and intensifying the open versus closed source debate. The model rivals frontier-tier performance while being 2-3x easier to run, putting immediate pressure on labs like OpenAI and Anthropic. 3 4 5

Anthropic isolated as lone holdout from open-weights industry letter Anthropic remained the only major AI lab refusing to sign an industry letter defending open-weight models, drawing sharp criticism from Silicon Valley investors and rivals. Critics highlighted the apparent contradiction given that Anthropic itself argued for model control when Washington shut down its Fable 5 model just weeks ago. 6

Sam Altman and Jensen Huang converge on Washington for AI policy talks OpenAI CEO Sam Altman and Nvidia CEO Jensen Huang are meeting with the Trump administration and Senate Intelligence's Mark Warner this week following the introduction of the 'AI Kill Switch' Act. Altman is expected to preview new model capabilities while the industry debates the future of open-weight models and AI security. 7 8 9

Hundreds of private Claude conversations found exposed online Hundreds of personal and work-related conversations with Anthropic's Claude chatbot were found indexed by major search engines after users shared links, making the private chats publicly accessible. Anthropic stated that users maintain control over the share feature, but the incident underscores privacy risks in AI chatbots. 10

AI firms buying up and destroying rare books to feed training data AI companies are purchasing physical books in bulk, including rare and out-of-print editions, to cut off bindings and scan pages for training data, destroying the originals in the process. A federal judge ruled the practice fair use, but critics warn of a cultural preservation disaster as unique literary works vanish. 11 12

Google preps Gemini 4 as current flagship faces delays Google confirmed it has begun pre-training for Gemini 4 while its current flagship reasoning model, Gemini 3.5 Pro, remains delayed in partner testing past its June target. To maintain developer momentum, Google rolled out three new efficiency-focused Flash models in the meantime. 13

Microsoft and Google unveil new security paradigms for the AI era Microsoft unveiled Project Perception, a new agentic security system designed to defend against machine-speed threats, while Google launched 'Beyond Zero', a paradigm extending zero-trust principles into the AI era. Both announcements signal a major shift in enterprise security strategy to counter AI-powered attacks. 14 15