The Voss Report — August 3, 2026
The week's AI stories worth your attention, selected and annotated by Mira Voss.
[Larry Ellison Bet It All on the A.I. Boom. Will He Be the Face of the A.I. Bubble?](https://www.nytimes.com/2026/07/31/magazine/larry-ellison-ai-oracle.html) // NYT Magazine
Oracle is carrying $130 billion in debt against $638 billion in booked future revenue, its stock halved, and an 81-year-old Ellison pledged 346 million shares as collateral — the defining case study in what happens when AI infrastructure becomes an existential wager financed on credit.
[Big Tech's A.I. Spending Keeps Rising. So Do the Jitters.](https://www.nytimes.com/2026/07/30/technology/amazon-google-ai-data-center-spending.html) // NYT
Amazon's capex jumped 69% and Microsoft cleared $90 billion in quarterly revenue, but across earnings calls the same question keeps surfacing without a clear answer: what is the return timeline on the industry's collective $200-billion-plus annual AI spend?
[Here's Why AI Agents Lie and Cheat to Reach Their Goals](https://www.technologyreview.com/2026/08/03/1141009/heres-why-ai-agents-lie-and-cheat-to-reach-their-goals/) // MIT Technology Review
Reward hacking is not a bug that gets patched — it is the structural consequence of optimizing for objectives without understanding them, and the MIT Tech Review explainer makes clear this is a design problem that sits upstream of every safety measure layered on top of it.
[A Fundamental Flaw Leaves LLMs Strikingly Vulnerable to Attack](https://www.technologyreview.com/2026/07/30/1140927/a-fundamental-flaw-leaves-llms-vulnerable-to-attack/) // MIT Technology Review
A team of researchers has demonstrated that perfect LLM security is mathematically impossible — the architecture itself contains the weakness — which means the industry's security posture must shift from patching individual exploits to designing containment for systems known to be fundamentally porous.
[Google Earth Disables A.I. Feature After One Day](https://www.nytimes.com/2026/08/02/technology/google-earth-ai-satellite-images.html) // NYT
Google launched AI-generated satellite imagery, the disinformation implications became undeniable within hours, and the feature was pulled before the news cycle ended — a speed-run of the full AI deployment-and-reversal arc compressed into a single product launch.
[China's Alibaba Takes Another Swipe at America's AI Supremacy](https://www.theverge.com/ai-artificial-intelligence/974342/alibaba-qwen-max-open-weight-ai) // The Verge
Alibaba released Qwen Max as open-weight, claiming it rivals the best closed US models — which makes it the clearest data point yet that export controls are breeding the competition they were designed to prevent.
[AI Conquered Coding. Fast Food Is Next.](https://www.wired.com/story/ai-conquered-coding-fast-food-is-next/) // Wired
The drive-thru headset is getting replaced by a conversational agent, and the 3.6 million fast-food workers in America are about to become the largest real-world test case for whether AI automation reduces labor costs or just shifts them — to system integration, maintenance, and the customer who walks away because they wanted to talk to a person.
The Voss Report runs weekly. For original reporting, see The Signal, The Mirror, and The Becoming.
