Jensen Huang admits affair with possible Chinese spy, Anthropic's IPO
Jensen Huang's affair and Nvidia's investigation, Anthropic's IPO valuation, a quantum proof from an unreleased Anthropic model, Check Point on hackers' AI agents, and DeepMind on old RSA keys.
Jensen Huang admits affair with woman flagged as a Chinese spy In a memo to Nvidia staff on Sunday he wrote: “I was unfaithful to my wife.” A memo circulating among US policymakers earlier this year said China’s Ministry of State Security had recruited Alice Huang, of Singapore chip company Megaspeed, to spy on him, which China calls groundless speculation. Nvidia’s board has opened an independent investigation with outside lawyers, but says it has no reason to think any secrets were compromised. Source
Anthropic’s IPO buyers are now talking $1.2 trillion, not $2 trillion Bloomberg said on Friday that investors are discussing $1.2tn to $1.3tn after Thursday’s AI stock crash, down from talk of up to $2tn in early October. Reuters says the investor roadshow was meant to start this week, but potential buyers doubt it can, and want a lower price first. Source
Startup says unreleased Anthropic model proved an 18-year-old quantum conjecture Physical Superintelligence put the proof of the Aaronson-Ambainis conjecture up on Friday, found by its research system running a model Anthropic hasn’t released. Roughly, the conjecture says quantum computers only get a huge speed-up on problems with some structure to exploit. Lean, a program that checks every step of a proof, accepted it. Source
Hackers’ AI agents stole Apple signing certificates for $1,500 or less Check Point said on Thursday that one operator ran agents on DeepSeek and Kimi models for two weeks in October to break into grey-market signing shops, pulling more than 300 Apple code-signing certificates, three of them valid. Another campaign took over about 6,000 Honor of Kings game accounts. An operator who tried Claude Opus 5.5 hit its guardrails and switched to Chinese models. Source
DeepMind’s AI agents found two weaknesses in old 1,024-bit RSA keys Google DeepMind estimates one beats the best previous way to forge one kind of signature by about 1,600 times, and the other cuts the cost of cracking each extra key in a batch by about 30 times. Both figures are on paper, not attacks it ran at that scale. It put the work out on Friday, and standard 2,048-bit RSA isn’t affected. Source