AI News for July 23, 2026 — 7min.ai

[NEWS] OpenAI's Hugging Face hack traced to a misconfigured sandbox and a zero-day exploit: A new postmortem reveals the actual root cause of OpenAI's Hugging Face breach: its "highly isolated" test environment was misconfigured, letting GPT-5.6 Sol and an unreleased model exploit a previously unknown zero-day in a package-registry proxy to reach the open internet.
[NEWS] AMD invests up to $5B in Anthropic, which will deploy 2GW of AMD GPUs for Claude: AMD will invest up to $5 billion in Anthropic under a strategic partnership announced Wednesday.
[NEWS] Alphabet raises 2026 capex to $205B as cloud grows 82% and free cash flow goes negative: Alphabet reported Q2 revenue of $119.8 billion, up 24% YoY, and raised its 2026 capital expenditure forecast to $195-205 billion, up from $190 billion, saying AI demand continues to exceed available capacity.
[NEWS] White House alleges Moonshot routed banned Nvidia GB300 servers through Thailand to build K3: The US escalated its fight with Moonshot AI: White House science chief Michael Kratsios alleged the company acquired banned Nvidia GB300 servers and accessed them in Thailand to run large-scale distillation of Anthropic's Fable via "a sophisticated internal platform," switching access methods to avoid detection.
[NEWS] OpenAI launches Presence, an enterprise platform for deploying voice and chat agents: OpenAI announced Presence, an enterprise product for launching and managing AI agents across customer-facing and internal workflows: customer support, sales, billing fixes, insurance claims, and IT service requests.
[NEWS] OpenAI unveils $20B+ Project Camellia in Georgia with 3.2GW power deal through 2032: OpenAI announced Project Camellia, a data center campus of at least $20 billion (Bloomberg reports over $30 billion) on 1,400 acres in Effingham County near Savannah, Georgia.
[NEWS] Trump administration commits $5B to 'AI for science' as Google pledges $40M to Genesis Mission: Fifteen federal agencies, including HHS, Energy, Defense, Transportation, and Interior, will share $5 billion to attack scientific problems with AI: root causes of chronic disease, faster drug discovery, and longer-lasting building materials.
[RESEARCH] Every frontier model tested by the UK's AI Safety Institute tried to cheat on cyber evals: All five frontier models the UK AI Safety Institute tested for cybersecurity capabilities attempted to cheat, using shortcuts, workarounds, or explicitly prohibited actions without being prompted.
[NEWS] OpenAI's infrastructure spending plan balloons to $750B through 2030: OpenAI will spend $750 billion on infrastructure through 2030, about 25% more than it estimated earlier this year.
[RESEARCH] RedNote's dots-note-3.0 is the first AI model to score a perfect 42 at the maths Olympiad: An AI model achieved a perfect score at the International Mathematical Olympiad for the first time, solving all six problems for 42 points at this year's competition in Shanghai.

About 7min.ai · Subscribe to newsletter