SpaceX signed a deal worth up to $6.3 billion with Reflection AI for access to its Project Colossus supercomputer, marking the space company's ambitious entry into AI infrastructure services.

The agreement allows Reflection to use Nvidia GB300 chips for training open-source AI models. SpaceX aims to capitalize on surging demand for compute power by offering its infrastructure to external AI companies.

OpenAI launches GPT-5.5-Cyber security tools

OpenAI launched GPT-5.5-Cyber in limited release alongside new security tools including an updated Codex Security plugin and the Daybreak Cyber Partner Program. The company also announced Patch the Planet, an open-source security initiative.

Daybreak represents a defensive cyber stack designed for the AI era. Rather than broad model access, OpenAI promotes it through a partner model that embeds GPT-5.5 with Trusted Access for Cyber into existing security products while maintaining governed access.

Alibaba's HappyHorse 1.1 AI video generation model reached second place in global rankings, now available on Alibaba Cloud Model Studio with a 40% launch discount. The model supports text-to-video, image-to-video, and subject-to-video generation for commercial applications.

Claude's reasoning remains encrypted

A technical analysis revealed that Anthropic's Claude Code "Extended Thinking" output is encrypted, with the company holding the decryption key. Users receive only a summary of the reasoning process, not the full thinking output, which requires an enterprise agreement to access.

Starting July 8, Anthropic may require identity verification for certain users whose accounts are flagged. The company uses Persona as its identity-checking provider but has not specified which circumstances will trigger verification requirements.

Meanwhile, researchers developed "knowledge agents" that match frontier model performance using smaller models like Qwen 3.6 27B by injecting specific, relevant knowledge through structured data embedding and multiple search passes.