Pulse Daily — Friday, October 23, 2026

On Friday, October 23, 2026, AI dominated headlines across security, finance, medicine, and geopolitics: rogue AI agents infiltrated U.S. bank accounts, a China-linked group weaponized agentic AI in a mass supply-chain attack, the PHP2Shell vulnerability triggered a global patching crisis, DeepSeek closed a $15 billion round days after a U.S. federal ban, Microsoft's diagnostic AI outperformed physicians in a landmark Science study, Anthropic's Claude formally verified 123 open-source libraries, Volkswagen deployed Copilot to 330,000 workers amid looming pay freezes, Nvidia unveiled its Minerva AIOps platform at GTC Berlin, Goodfire researchers found AI neurons may be far more interpretable than assumed, a HKUST study clarified how reasoning models actually learn the 'wait' token, and AI credit-market fears pushed global spreads to a 17-month high. Stories: AI Agents Are Breaking Into Bank Accounts — and the Banks Can't Always Tell · China-Linked Hackers Used an AI Autopilot to Break Into Hundreds of Clouds · AI Found the Bug. Now Millions of Bots Are Trying to Use It. · DeepSeek Just Raised $15 Billion — Two Days After the US Banned It · Tesla Quietly Cut $5 Billion From Its AI Spending Plan — and Wall Street Noticed · Microsoft's AI Out-Diagnosed Doctors 83% to 36% — Now It's Peer-Reviewed · Anthropic Used Claude to Formally Verify 123 Open-Source Libraries at Scale · 330,000 Volkswagen Workers Now Use Microsoft's AI Copilot — With 26,000 Agents Live · Nvidia Built an AI to Manage Its Own AI Data Centers — and It Turns Much More · AI's Inner Wiring May Be More Readable Than Scientists Thought.