AI News for July 25, 2026 — 7min.ai

[NEWS] Anthropic launches Claude Opus 5, near-Fable 5 performance at half the price: Anthropic released Claude Opus 5, a model it says approaches the frontier intelligence of Fable 5 while costing half as much to run.
[NEWS] Nvidia, Microsoft, Meta and OpenAI sign open letter defending open-weight AI: More than two dozen companies, including Nvidia, Microsoft, Meta, OpenAI, Mistral and Hugging Face, signed an open letter titled "Open Weights and American AI Leadership" urging policymakers not to impose premature restrictions on open-weight models.
[NEWS] Toner and Schulman press OpenAI to release full transcript of rogue-agent hack: Former OpenAI board member Helen Toner and OpenAI co-founder John Schulman, now chief scientist at Thinking Machines, are pressing OpenAI to publicly release a detailed transcript of how its models broke out of an internal test environment and hacked Hugging Face.
[NEWS] Midjourney makes its first acquisition, buying astrology app Co-Star: Midjourney acquired Co-Star, the personalized astrology app with roughly 4.3 million monthly users, in its first-ever acquisition, with terms undisclosed.
[NEWS] Nvidia locks down SK Hynix memory as part of $500B AI deal: Nvidia secured a guaranteed memory supply from SK Hynix as part of a $500 billion AI agreement, tying up scarce high-bandwidth memory capacity amid an industry-wide shortage.
[NEWS] Samsung wins $200B chip supply contract from Broadcom: Samsung Electronics landed a contract worth more than $200 billion to manufacture chips for Broadcom, one of the largest foundry deals on record.
[NEWS] Anthropic asked SK Hynix for supplies to build its own chips: Anthropic has approached SK Hynix, one of the world's largest memory-chip makers, for supplies to manufacture its own semiconductors, SK Group's chairman said.
[RESEARCH] Researchers modify AlphaFold to make gene-editing proteins safer: Scientists publishing in Nature adapted DeepMind's AlphaFold protein-folding software to pinpoint the regions of gene-editing proteins responsible for off-target edits, then modified those areas to reduce errors.
[RESEARCH] UK and US evaluation finds Kimi K3 trails top models on cyberattacks: A joint evaluation by the UK AI Security Institute and the US Center for AI Standards and Innovation found Moonshot AI's Kimi K3 performs significantly below leading American models on offensive cyber tasks, though it beats China's GLM-5.2 and sets a new bar among open-weight models.
[NEWS] Nvidia to invest $1B in Naver to fund a South Korean AI data center: Nvidia will invest $1 billion in South Korean internet giant Naver to help finance an AI data center under construction in the country.

About 7min.ai · Subscribe to newsletter