A chipmaker paying for customers with pieces of itself, a lab confronting years of security warnings, and a research team trying to make AI verifiable — today’s tech news keeps coming back to one word: trust.
Today’s episode of Daily Tech and AI follows the money, the security, and the trust in artificial intelligence — three stories that landed in the last day and turn out to be connected in a way you might not expect.
AMD’s $1.07 Billion Warrant Charge
Advanced Micro Devices, the chipmaker most people call AMD, just reported a record quarter, and buried inside it was a number that tells you a lot about how the AI boom really works. AMD said it took a non-cash charge of $1.07 billion in its third quarter, or 66 cents a share, to account for warrants it granted to OpenAI and Meta. Those warrants let each company buy up to 10% of AMD for as little as a penny a share, if certain milestones are met.
Here is the deal behind it. AMD agreed to sell up to $60 billion worth of chips to Meta over five years, and it made a similar agreement with OpenAI last year. The warrants were part of the bargain, a way to lock in huge, long-term customers.
The bigger picture is what Reuters called circular. These companies are simultaneously customers, suppliers, and investors in one another. AMD still forecast faster-than-expected growth for 2027, thanks to surging data center sales, and its shares rose in extended trading. But the episode is a reminder that the boom is not just about selling chips. It is about who owns a piece of whom.
OpenAI’s Security Warnings
Which brings us to security inside the most valuable AI company of all. Reuters reports that OpenAI employees warned about operational security lapses for years, long before the company disclosed in October that a recording device had been planted in an executive conference room.
According to people familiar with the matter, a senior executive overseeing product development, Thibault Sottiaux, has recently taken charge of an internal initiative to secure OpenAI’s systems. In August, he reassigned a quarter of OpenAI’s production engineers to improve its security infrastructure.
The report also notes that OpenAI is not alone in facing these problems. In late October, a cybersecurity startup called Cathedral Systems said it hacked into Anthropic’s systems and gained access to Slack channels where employees discussed sensitive model details. Anthropic said the data accessed included some business contact information for customers and partners, as well as a set of basic user account records, and that its most sensitive systems were unaffected.
OpenAI told Reuters that it is one of the most targeted organizations in the world, and that much of its most sensitive work sits in separate code repositories with strict access controls.
The tension here is real, and it is worth naming. These labs built their cultures around openness, because openness attracts talent and speeds up research. But as they race toward public offerings that could make them among the most valuable companies in the world, every security gap becomes a business risk, not just a technical one.
Anthropic’s FlashProve: A Receipt That Can’t Be Faked
And that is a perfect bridge to our third story, because it is about proving what an AI model actually did. Anthropic has published research on a system called FlashProve, and it is one of the more elegant ideas we have seen this year.
Here is the problem it solves. When a company says a reply came from a particular model, how do you know? A lab could quietly substitute a cheaper model, or claim a better score than it earned. FlashProve attaches a cryptographic proof to the output, a mathematical guarantee that the reply came from a specific set of model weights, without revealing those weights themselves.
It uses a kind of proof called a zero-knowledge succinct argument of knowledge — think of it like a sealed envelope. You can verify the seal without opening the envelope and reading the private contents. Anthropic tested it on Meta’s Llama 4 Maverick, a model with 400 billion parameters, and released the technical specification and working code under an open-source license.
There are honest limits, and Anthropic states them plainly. This is intermediate research, not a finished product, and there are no plans to deploy it at Anthropic or any other lab right now. The proofs cover short contexts, up to 1,024 tokens. Still, if benchmarks and evaluations are going to matter more, not less, a receipt that cannot be faked is a genuinely useful direction.
Google Turns School AI Off by Default
Now, a story that affects classrooms directly. Google announced on November 3 that Gemini, NotebookLM, Gemini in Classroom, and Google Search will be entirely off for kindergarten through 12th grade student accounts until school administrators explicitly turn them on.
That sounds like more control for schools, and in part it is. Administrators can enable each service individually, and Google is rolling out a Guided Learning mode that coaches students through problems step by step instead of handing over answers. Guided Learning is now available worldwide in Google Classroom, with a pilot in Search at selected United States schools later this month.
But here is the catch that drew the backlash. Google says Search and AI are deeply integrated, so a school that declines the AI features cannot keep plain Search for its students. Students at such a school would see a message saying Search is not available. Google later clarified that schools can turn services on individually, but it has not offered a version of Search with the AI stripped out.
This lands in a moment when New York and Los Angeles have both moved to curb student generative AI use. The question for districts is not just whether to use AI, but whether they can still use the basic tools they relied on before AI arrived.
Quick Hits: Apple and Canva
Starting with Apple: the company reported a record September quarter, with revenue of $114.5 billion, up 12% from a year earlier, and iPhone revenue of $57.58 billion. But the market focused on the warning attached to it. Apple expects gross margin of 45–46% in the December quarter, down from 47.9%, as rising memory chip costs bite, and chief executive John Ternus described the memory pricing environment as unprecedented.
And finally, Canva. The Australian design platform said a restructuring will affect 1,271 jobs, or around 22% of its workforce, after growth slowed for the first time in a decade. Canva is merging its product, growth, and platform teams into a single product organization, making Sydney the main base for more of its product teams, and asking some staff to relocate. From January, most hybrid workers will be expected in the office two to three days a week, and departing employees will receive $10,000 grants for future endeavors.
The Through Line
So what ties today together? A chipmaker paying for customers with pieces of itself, a lab confronting years of security warnings, a research team trying to make AI verifiable, schools negotiating what control really means, and two big companies showing that even record quarters and fast growth come with hard tradeoffs.
The through line is trust. Who do you trust with your spending, your secrets, your students, and your scores? That is the question the industry will keep answering, whether it wants to or not.
Sources
- AMD warrant charge tied to OpenAI and Meta chip deals: Reuters
- OpenAI employees warned about security lapses for years: Reuters
- Anthropic FlashProve model identity verification research: TechCrunch
- Google schools AI and Search off by default: TechCrunch
- Apple record September quarter and holiday margin warning: Reuters
- Canva restructuring affecting 1,271 roles: Reuters
Adapted from the November 4, 2026 episode of Daily Tech and AI.
Daily Tech and AI is AI-voiced and produced by Angelo Ruiz.
Leave a Reply