Microsoft's MAI Models Found Trained on Unlicensed Web Data

A major contradiction has emerged in the training methodology of Microsoft's own LLM, which the company had promoted as different from its competitors. Microsoft claimed that its new MAI models were trained using only enterprise-grade, clean, commercially licensed data, but reports reveal that unlicensed web data such as Common Crawl was actually used in training.

Like other AI companies, Microsoft reportedly justifies its use of unlicensed data on "fair use" grounds, effectively shifting the burden of blocking crawlers onto website operators. Transparency and copyright issues in AI companies' data collection practices have become an industry-wide challenge, and the gap between Microsoft's "clean" claims and the reality deserves attention.

Anthropic's Mythos Repurposed for NSA Offensive Cyber Operations

Reports indicate that Anthropic's AI model Mythos is being used in offensive cyber operations by the U.S. National Security Agency (NSA) targeting China and Iran. Anthropic has reportedly stationed about six engineers directly at the NSA to adapt Mythos for offensive use.

While the company has publicly pledged restrictions on AI use, constraints such as those against mass surveillance explicitly apply "only to U.S. citizens." Where the line should be drawn for AI companies cooperating with security agencies is likely to spark international debate.

AirTrunk to Build AI Data Centers in India with $30 Billion

AirTrunk, an Australian data center operator, has announced a $30 billion (approximately 4.5 trillion yen) investment to build 5 GW (gigawatts) of AI data center capacity in India. This is the latest example of large-scale infrastructure investment responding to the surge in AI workloads, signaling India's growing importance as a data center market.

Chinese-Made PCBs in Nvidia AI Chips Raise Security Concerns

The revelation that printed circuit boards (PCBs) used in Nvidia's AI chips are made in China is raising security concerns in the United States. This once again highlights the risks of China dependence in high-tech supply chains. The argument that even U.S.-made chips can carry security vulnerabilities through the provenance of peripheral components could have ripple effects across the entire AI semiconductor industry.

Stanford HAI: AI Hiring Tools May Produce Racial Bias

Stanford's Institute for Human-Centered Artificial Intelligence (HAI) has released research findings indicating that AI hiring tools may produce racial bias and lead to systematic rejections of candidates. As AI-driven hiring automation spreads, ensuring fairness has become an urgent priority. The study emphasized the need for transparency in AI tools' decision criteria and for continuous auditing.

EU to Ask Households to Cut Electricity Use Amid Rising AI Demand

The EU has revealed plans to ask households to reduce electricity consumption in order to cope with surging industrial and AI demand. As AI data centers consume enormous amounts of power, a new fault line is emerging over electricity allocation between the residential sector and industry. The strain that AI growth places on social infrastructure is already reaching a level that can no longer be ignored.

---