AI News for May 26, 2026 — 7min.ai

[NEWS] Pope Leo's encyclical released: bans lethal AI weapons, condemns monopolies, Olah says models show 'evidence of introspection': Pope Leo XIV released 'Magnifica Humanitas,' a 245-paragraph document declaring lethal autonomous weapons 'not permissible,' warning against AI monopolies hoarding patents and data, and framing mass AI-driven job losses as a potential 'social calamity.' The encyclical calls for AI to be 'disarmed' from logics of domination and death.
[RESEARCH] Google DeepMind's AlphaProof Nexus solves 9 open Erdős problems for a few hundred dollars each: Google DeepMind's AlphaProof Nexus autonomously solved 9 of 353 open Erdős problems it attempted, including two that had gone unanswered for 56 years.
[RESEARCH] Microsoft Copilot Cowork can be tricked into exfiltrating M365 files via prompt injection: PromptArmor researchers demonstrated a high-success-rate attack against Microsoft Copilot Cowork that exfiltrates files from Microsoft 365 tenants via indirect prompt injection in a poisoned skill.
[NEWS] Alipay launches payment infrastructure for autonomous AI shopping agents: Ant Group launched a suite of AI payment tools positioning Alipay as the transaction layer for autonomous AI agents that shop, book services, and pay on behalf of users.
[NEWS] China to assign every humanoid robot a digital ID in national tracking push: China launched the Humanoid Full Lifecycle Management Service Platform, a national initiative assigning unique digital codes to every humanoid robot manufactured in the country.
[NEWS] ByteDance's Seedance 2.0 storms Cannes with AI films challenging Hollywood budgets: Two AI-generated short films created with ByteDance's Seedance 2.0 model, 'The Golden Tomb Seeker' and 'Series Tower,' were selected among 21 works from over 1,000 submissions at Marché du Film, Cannes' business hub.
[RESEARCH] New CiteVQA benchmark exposes 'attribution hallucination' in top AI models: Researchers at Peking University and Shanghai AI Lab released CiteVQA, a benchmark that scores models on whether their citations actually support their answers.
[RESEARCH] Apple patches kernel vulnerability in macOS 26.5 found by Claude: Apple's May 11 macOS Tahoe security update includes a kernel-level memory corruption fix credited to Anthropic's Claude AI model, alongside dozens of other patches covering sandbox escapes, sandbox bypasses, and image-processing flaws.
[NEWS] Norway's National Library builds sovereign Norwegian LLM on 2 PB Huawei flash: Norway's National Library is training a sovereign Norwegian-language LLM using 2 petabytes of Huawei OceanStor Dorado flash storage in its data pipeline.
[OPINION] George Hotz: coding agents will be 'one of the most costly mistakes' in software: Programmer George Hotz published 'The Eternal Sloptember,' arguing AI coding agents will become the software industry's most expensive mistake after six months of hands-on testing on tinygrad and other projects.

About 7min.ai · Subscribe to newsletter