AI News for July 28, 2026 — 7min.ai

[TOOLS] Microsoft launches MAI-Cyber-1-Flash, claims it beats frontier models at half the cost: Microsoft unveiled its first custom-built cybersecurity model, MAI-Cyber-1-Flash, embedded in its MDASH multi-agent harness for finding and fixing software vulnerabilities.
[NEWS] Nvidia pours billions into Ilya Sutskever's Safe Superintelligence: Nvidia announced a long-term strategic partnership with Safe Superintelligence, the secretive lab founded by former OpenAI chief scientist Ilya Sutskever, alongside an investment that stretches into multiple billions.
[RESEARCH] Moonshot's Kimi K3 launch reveals 104B active params, custom license restrictions: Moonshot AI followed through on its planned Kimi K3 release, publishing the complete weights alongside inference infrastructure, optimized attention kernels, MoE communication libraries, and a 47-page technical report.
[NEWS] Anthropic explains Claude chat leak, but exposed links stay live on Bing and Brave: Anthropic addressed the exposure of shared Claude conversations in Google search results, saying it does not submit chat directories or sitemaps to search engines and that shared links aren't guessable, attributing the leak to third-party archiving of publicly accessible pages once shared.
[NEWS] Apple seeks US permission to buy CXMT memory as PC makers weigh the blacklisted chip: Apple has asked the Trump administration for permission to purchase memory from CXMT, China's DRAM maker whose Shanghai debut this week made it the most valuable listed Chinese company.
[NEWS] Amodei says Anthropic never advocated banning open-weight models: Dario Amodei published a blog post rejecting speculation that Anthropic wants Washington to ban open-weight AI models.
[NEWS] Nvidia, Microsoft, SpaceX form Open Secure AI Alliance without OpenAI, Google, or Anthropic: Nvidia unveiled the Open Secure AI Alliance, a coalition to build and share open-source AI security tools in direct response to OpenAI's rogue agent breaching Hugging Face.
[RESEARCH] OpenAI research finds workers using ChatGPT to do other people's jobs: OpenAI published research on what it calls "task crossover," finding that 43.5% of job-specific ChatGPT queries involved work outside the user's own profession.
[NEWS] Apple retakes most-valuable-company crown as Nvidia's $750B deals stir bubble fears: Nvidia stock fell nearly 5% on Monday, dropping its market cap to about $4.77 trillion and letting Apple reclaim the title of world's most valuable company at close to $5 trillion.
[RESEARCH] Epoch and METR release MirrorCode, showing AI rebuilds week-long software projects: Epoch AI and METR released MirrorCode, a benchmark that tasks AI agents with reimplementing entire command-line programs from behavior alone, without source code or web access.

About 7min.ai · Subscribe to newsletter