Welcome to the newest edition of ID Tech’s AI Update. Here’s the latest big news on the shifting landscape of AI and identity technology:
Meta has turned to a massive bond issuance to support its planned AI expenditures, selling $30 billion of investment-grade bonds across six tranches with maturities from 5 to 40 years. The move came after a ~10 percent drop in Meta stock as shareholders recoiled from plans for a $70–$72B capex this year and “notably larger” capex in 2026. “I think that it’s the right strategy to aggressively front-load building capacity, so that way we’re prepared for the most optimistic cases,” said CEO Mark Zuckerberg.
OpenAI has restructured, helping to turn Microsoft into a $4 trillion company. A long-negotiated recapitalization agreement puts a nonprofit foundation in control of a new for-profit public benefit company, OpenAI Group PBC. Microsoft, OpenAI’s biggest investor, gets 27 percent of the for-profit arm – a stake valued near $135 billion at a $500 billion OpenAI valuation. The move is reportedly part of OpenAI’s effort to lay the groundwork for a potential IPO that could target a valuation as high as $1 trillion.
Samsung will buy about 50,000 Nvidia GPUs to power a new “AI factory” for chip manufacturing, devices, and robotics. It’s part of Nvidia’s larger dealmaking effort in South Korea, with the firm planning a roughly 260,000-GPU rollout that will see major allocations (about 50,000 GPUs) to SK Group and Hyundai, and more chips earmarked for sovereign AI compute via the National AI Computing Center and cloud providers. CEO Jensen Huang was touring the country this week, and also expressed hope that someday US export policies would allow his firm to sell chips to China, too.
Character.AI will ban users under 18 from open-ended chats with its AI-powered characters starting November 25, and has immediately restricted teen chat time to two hours per day. The company says that existing teen accounts will be redirected to a teen-safe experience during the transition. It’s rolling out an in-house age-assurance model (which looks at character choices and other on-site and third-party signals), and will combine it with Persona for ID verification if adults are misclassified.
A deepfaked version of Jensen appeared in a fake Nvidia livestream that ran concurrently with the real “Nvidia GTC” conference livestream on YouTube this week. It was a crypto scam, with the fake Jensen urging viewers to scan a QR code presented on screen and transfer funds to support a project tied to “Nvidia’s mission to accelerate human progress.” The fake livestream reportedly had something like 95k-120k concurrent viewers – about five times the real keynote’s live audience.
A growing body of research shows that voices can be algorithmically derived from photos of people’s faces and used to fool voice-recognition systems. A major paper by researchers at the National University of Singapore and KAIST described a Face-to-Voice (FOICE) attack that generated synthetic speech from a single face image and bypassed WeChat Voiceprint roughly 30 percent of the time on a first try, approaching 100 percent when attackers were allowed repeated attempts. This year, researchers at Australia’s CSIRO and Macquarie University published follow-up tests showing that state-of-the-art audio deepfake detectors “consistently fail” to detect Face-to-Voice audio under both clean and noisy conditions.
The chatbot’s take: GPT-5 breaks down the FOICE science.
–
By Alex Perala
Follow Us