Verified reporting and clear context on what is changing in artificial intelligence — and what remains a promise.
Five signals to understand the edition in under a minute.
Image: official source / editorial archiveNvidia, Microsoft, IBM, the Linux Foundation and 33 other organizations have formed an alliance to share agent-security tools. OpenAI, Google and Anthropic are not founding partners.
Read the full analysisOriginal reporting based on primary sources, with no inflated headlines or automated summaries.
Image: official source / editorial archiveNvidia, Microsoft, IBM, the Linux Foundation and 33 other organizations have formed an alliance to share agent-security tools. OpenAI, Google and Anthropic are not founding partners.
Image: official source / editorial archiveDario Amodei rejects a blanket ban and instead calls for controls on advanced chips, industrial-scale distillation and mandatory testing for the most capable models.
Image: official source / editorial archiveNvidia and Safe Superintelligence confirm a partnership and “substantial” investment. Reuters and the Financial Times put the equity investment at $5 billion; the announcement does not disclose the figure.
Image: official source / editorial archivePresence brings voice and chat agents into enterprise workflows with permissions, evaluations and human escalation. On July 29, OpenAI published GPT‑5.6 and research updates, but did not unveil another flagship model.
Image: official source / editorial archiveMoonshot AI’s model combines 2.8 trillion parameters, a one-million-token context and native vision. Fireworks and Ollama now offer confirmed access; Cursor has not officially announced it.
Image: official source / editorial archiveThe 2026-07-28 specification redesigns the protocol to scale on ordinary HTTP, hardens OAuth and makes interfaces and long-running tasks official extensions.
Image: official source / editorial archiveThe platform turns plain-language instructions into branching, scheduled workflows with human approval, model selection and cost controls.
Image: official source / editorial archiveThe new service connects existing tools and routes routine work to open models, with centralized observability and team controls.
Image: official source / editorial archiveAI Gateway can pin inference to the US or European Union and now supports persistent connections for long OpenAI Responses workflows.
Image: official source / editorial archiveThe lab released PerceptionBench, 3,000 questions separating perception from reasoning, and highlights FlashKDA, whose open kernels have existed since April.
Complete, original templates that are easy to adapt to real work.
Seven reviewed files with real sizes and direct, account-free downloads.
Open resources →Nvidia, Microsoft and other technology companies have formed the Open Secure AI Alliance to share open security tools. Anthropic says it does not want open-weight models banned, but supports tighter chip controls, action against industrial-scale distillation and testing for the most capable systems. Nvidia is also expanding its infrastructure bet with a reported $5 billion investment in Safe Superintelligence. And OpenAI is bringing agents into enterprises with Presence, while its July 29 updates cover GPT‑5.6 and research rather than another flagship model.
Editable in the private Nexus AI dashboard.