CRUNCHY TECH
YESTERDAY, STILL WORTH KNOWING
Nvidia's board investigates its CEO, a Linux flaw comes under active attack, China's giant AI campus advances, and Anthropic, Disney, DeepMind, Waymo and Iris Plaza all make news.
Nvidia, the US chipmaker whose processors are viewed as crucial to American technological supremacy, has opened an independent board investigation into reports that CEO Jensen Huang was targeted by an alleged Chinese intelligence operation through an extramarital affair. In a memo to staff, Huang wrote: 'I owe my family and all of you an apology. I was unfaithful to my wife. My actions hurt her and our children. There is no excuse for what I did.'
The claims, reported by The New York Times and the Financial Times, concern Alice Huang, an executive at chip buyer Megaspeed, whom US agencies reportedly believe was recruited by China's Ministry of State Security. China dismisses the espionage claims as unfounded, and the operation has not been independently confirmed. A committee of independent directors, assisted by outside counsel, will oversee the inquiry. The board says it 'has no reason to believe that either national security or NVIDIA secrets were compromised', has confidence in Huang, and is not seeking his removal. The stakes are high because Megaspeed is reportedly under US Commerce Department investigation over whether it smuggled AI chips to China, and because Huang's closeness to President Trump raises questions about what insights into the administration's AI deliberations he may have held.
A critical vulnerability nicknamed *sshole is now under active attack. It exploits the interaction between OpenSSH, the standard tool for secure remote logins, and musl, a lightweight code library some Linux systems use instead of the more common glibc, allowing unauthenticated access to high-privilege accounts within seconds. Security firm Greynoise says honeypots it operates in Southeast Asia have been targeted by a group it tracks as GT10, describing the intrusions as 'mature intrusions'; its assessment is that the deployed backdoor contains Chinese-language comments and appears AI-generated.
Google DeepMind says AI agents, working with cryptographers from UC San Diego and Inria, found cheaper attacks on 1024-bit RSA, a legacy encryption standard still found in older systems. In a sandboxed test they cut a forgery's computing cost 3.5-fold, and produced a paper algorithm that would make each extra key around 30 times cheaper to factor on average. The attacks do not recover private keys, and DeepMind's Pushmeet Kohli said modern protocols using 2048-bit or larger keys remain secure, though he argued cryptographic hardness 'can no longer be treated as a fixed assumption'. Separately, Japanese homeware maker Iris Ohyama has taken its Iris Plaza shopping site offline after hackers may have accessed personal data of 7.7 million people, including names, addresses and passwords. It says no credit card data was stored there, but it is still unclear whether any data was actually taken.
China is building an 800MW AI data centre campus in Inner Mongolia that a SemiAnalysis report says could undercut America's Stargate project on cost. Envision has completed the first 120MW in under a year at roughly one-fifth of comparable US costs, powering it with on-site wind farms and batteries rather than the grid. US export restrictions on advanced chips remain a separate constraint.
Anthropic, the AI company behind Claude, will pay $3.4 billion for 1.7 gigawatts of gas generators from Caterpillar orders originally placed by supplier OnePWR Solutions, Reuters reported. With turbine wait times reportedly running to six years, the deal skips the queue but has drawn environmental criticism.
NVIDIA says it is working with Korean firms to move physical AI, meaning AI that operates in the real world, into production: Samsung on chip design, where one reported simulation case was cut from about 18 hours to 41 seconds, Kakao Mobility on level 4 autonomous driving, and LG Electronics on factory robots.
Disney and Adobe have announced a multiyear partnership, unveiled at Adobe's MAX conference, to build custom, artist-led generative AI models trained on Disney's own material for its studios, ESPN, theme parks and cruise line. It follows a mixed record, including a previously abandoned OpenAI deal. Meanwhile, US regulator NHTSA has opened a recall query into Waymo robotaxis after they kept illegally passing stopped school buses even after a software fix Waymo said was completed on November 17, 2025. Waymo says it has made significant improvements and received no citations since March 2026.