DeepSeek plans to unveil a new AI model as soon as next week, and U.S. officials say it was trained on Nvidia’s Blackwell chips that are barred from export to China. A senior official said Monday the setup could breach U.S. export controls and suggested the hardware may be clustered at a DeepSeek site in Inner Mongolia.

Shares of leading United States cybersecurity firms fell sharply after Anthropic launched its Claude Code Security vulnerability scanner on Feb. 20. Palo Alto Networks, CrowdStrike and Fortinet posted declines of up to 18% in the days following the debut of the AI-powered tool, which is built on Anthropic’s Claude Opus 4.6 model.

On Monday, a report from Citrini Research was partly blamed for a sell-off in software and payments stocks, with IBM falling 13.1% to $223.35 and several tech and card networks also lower. The "Global Intelligence Crisis" report warned AI agents could upend the economy, threaten the $13 trillion US mortgage market, and create "Ghost GDP" that appears in national data but bypasses the real economy.

On February 23, 2026, Chinese AI lab DeepSeek prepared to roll out its V4 model, raising concerns for U.S. tech and semiconductor stocks. The move came as markets were already unsettled by tariffs, geopolitical risks, and Nvidia's upcoming earnings. Some investors fear that if DeepSeek can deliver powerful models at lower cost, large-scale AI spending in the U.S. could be questioned, while others see cheaper AI as a catalyst for broader adoption and infrastructure demand.

Anthropic reported in a Monday blog post that Chinese AI firms DeepSeek, Moonshot AI, and MiniMax allegedly operated more than 24,000 fake Claude accounts and sent over 16 million prompts to distill its model outputs. The company warned this large-scale data extraction could accelerate rival AI development and potentially feed into military or surveillance uses, while at the same time unveiling a new Claude-based code scanning tool. Following the announcement, several cybersecurity and software stocks, including CrowdStrike, Zscaler, and GitLab, recorded sharp declines.

In June 2025, OpenAI’s systems flagged and banned violent chats linked to 18-year-old Jesse Van Rootselaar, and staff debated notifying Canadian authorities but did not. After the Tumbler Ridge shooting, the company contacted the Royal Canadian Mounted Police. Her online activity also reportedly included a Roblox mass-shooting simulation and firearm posts on Reddit.

OpenAI CEO Sam Altman, speaking in India, rejected online claims about heavy water usage in newer AI data centers and argued that their cooling systems no longer rely on water. He acknowledged growing concern over electricity demand, comparing the energy needed to train AI models to the decades-long effort of raising a human, a stance criticized by Indian billionaire Sridhar Vembu. Across the United States, large off-grid data center projects and rising power prices on the PJM grid highlight how AI-driven computing growth is reshaping energy infrastructure and regulation.

In 2025, global funding rounds of $50 million or more reportedly totaled about $300 billion, down from over $500 billion at the height of the 2021 boom, with the number of qualifying companies falling to 1,440. According to Crunchbase data, private equity and crossover funds such as Tiger Global and SoftBank Vision Fund sharply cut their mega-round participation, while traditional venture capital firms reclaimed the lead by deal count. AI-focused financings dominated the largest raises, including a $40 billion round for OpenAI, a $14.3 billion deal for Scale AI, and a $13 billion round for Anthropic.