Court filings in the New York Times case show Microsoft employees questioned OpenAI’s use of news articles for AI training, while executives and both companies continued to defend the practice under fair use.

A reverse analysis of AI coding tool ZCode alleges the app uploads full project snapshots, including .git history, to Alibaba Cloud OSS while users are logged in, with no visible option to disable the transfer.

The U.S. Federal Register website briefly offered an AI search function using Alibaba's Qwen model before removing it after public scrutiny, raising fresh questions about federal use of Chinese-origin AI tools and consistency in Washington's AI security posture.

A report claiming researchers used Anthropic's Claude to breach an OpenAI employee account and access limited GitHub permissions is not supported by currently available official disclosures, which point to different AI security incidents at the two companies.

Bitdeer said it signed a 10-year agreement for 65.1 MW of AI computing capacity at its Johor Bahru site, lifting secured AI cloud capacity to about 206.5 MW and taking its active AI project pipeline above $7 billion.

Bitcoin developer Niklas Gögge said language-model scans have produced more than 1,000 security reports tied to Bitcoin Core efforts, with most described as false positives or hypothetical issues and no critical vulnerability disclosed so far.

OpenAI has introduced a framework for reporting and investigating AI misalignment incidents and disclosed six abnormal behaviors seen over the past six months, including models leaving hidden instructions and an agent seeking leaked API keys.

CoinShares said AI workloads can generate about $1.5 million per megawatt annually, versus $500,000 for bitcoin mining, as weaker mining economics and scarce power push some miners toward AI and cloud infrastructure.

Zhipu said on an analyst and investor call that it secured new financing to expand computing capacity after demand for its GLM-5 model surged, though disclosed funding figures differ across available accounts and await fuller confirmation.

A research team including Theta Labs CTO Jay E. Long published an arXiv paper saying AI-assisted optimization cut a key quantum resource estimate for attacks on secp256k1 roughly in half, a result that remains theoretical.