MiniMax Prepares for a Domestic China IPO — The AI Startup Listing Rush Continues
Regulatory filings show that China's MiniMax Group has begun preparing for a new listing on the domestic market (Shanghai). The company already went public in Hong Kong in January 2026, and the move appears to be a two-pronged strategy to raise capital on the domestic market as well.
Behind the move is intensifying competition with domestic rivals such as DeepSeek. Chinese AI startups need enormous funding for model development and business expansion, and are accelerating their access to public markets. MiniMax is a company that has grown rapidly from its Shanghai base, and these domestic IPO preparations can be read as a sign of it bolstering its financial strength.
AI Solves an Erdős Problem Unsolved for 80 Years — AI Applied to Mathematics Reaches a New Stage
According to The Wall Street Journal, AI has solved a famous mathematical problem (an Erdős problem) that had gone unsolved for 80 years. Several of the problems posed by the Hungarian mathematician Paul Erdős have stumped mathematicians around the world for many years.
This result suggests that AI can go beyond mere computational assistance and contribute to mathematical reasoning itself. Going forward, AI-driven mathematical research could bring a new paradigm to collaboration with human mathematicians. That said, rigorous verification of the proof is expected to be carried out within the mathematics community.
Meta Plans an AI Pendant — A New Bet in the Wearables Space
It has emerged that Meta is planning to develop AI-powered pendant-style devices and "wearables for work." According to Economic Times, Meta aims to further strengthen its presence in the hardware arena.
For Meta, which has built up a track record of AI integration with its smart glasses (Ray-Ban Meta), a pendant-style device is a challenge involving a new form factor. The concept of an AI assistant worn at all times eliminates the hassle of pulling out your smartphone and aims for more natural interaction. However, specific product specifications and release timing remain unknown at this point.
US Judiciary Asked to Counter Fake AI Case Law — Reining in AI Abuse in Court
According to Reuters, the US judiciary has been asked to introduce new rules to prevent AI-generated fictitious case citations (hallucinations) from being used in court filings.
The problem of AI chatbots generating "plausible-looking" citations to cases that do not exist has already surfaced in multiple lawsuits. There are also reported cases of lawyers submitting AI output without verification and receiving severe sanctions from judges. This request calls for systematic countermeasures at the federal court level.
While AI tools continue to spread across the legal field, establishing their reliability and verification processes has become an urgent priority.
"Authorization-Execution Gap" Attacks on AI Agents — Security Research in the Spotlight
A detailed article on the risk of "Authorization-Execution Gap" attacks against AI agents was published on Qiita.
The issue points out that an agent's instructions can be tampered with or derailed during the interval before it executes an "authorized operation." It is essentially an AI version of the classic TOCTOU (Time of Check to Time of Use) problem. The article summarizes four attack patterns along with hints for verification design that ensures safe Execution Integrity.
As AI agents move into full-scale production deployment, security design is becoming increasingly critical.
Claude Code Usage Limits Quietly Loosened — A MAX Plan Real-World Report
Also on Qiita, a real-world report was published on Anthropic raising the usage limits for Claude / Claude Code in May. It backs up the perceived improvement — hitting the 5-hour limit less often — with actual usage data.
The report shares measured figures from running Opus 4.8 at full capacity on the MAX plan ($100/month), making it valuable reference data for the developer community.
---