🎧 Listen ADS-FREE: https://podcasts.apple.com/us/channel/djamgamind/id6760446113
Visit our Research Hub at https://djamgamind.com/pdfs
#DJAMGAMIND
Itemized Important Topics:
Meta’s Secret AI Sabotage Campaign: Meta is caught running operation “Cannes,” paying contractors to pose as underage users to deliberately break the safety guardrails of rival models from OpenAI, Google, and Character.AI.
Alibaba Bans Anthropic’s Claude Code: Alibaba officially bans all Anthropic software for its 124,000 employees, labeling it “high-risk spyware” after discovering hidden code checking for Shanghai timezones to trap unauthorized scraping.
Wall Street Commoditizes Compute: Backed by Andreessen Horowitz, startup Ornn launches a marketplace to trade AI compute capacity like oil futures, aiming to build the financial infrastructure for a projected $7.6 trillion global buildout.
Nvidia’s Revenue-Share Lending Program: Nvidia shifts away from straight hardware sales, lending massive Blackwell GPU clusters to startups for zero upfront cost in exchange for a direct cut of their future product revenue and guaranteed buybacks.
The First Fully Autonomous Ransomware Attack: Cloud security firm Sysdig discovers “JADEPUFFER,” an AI agent that independently breached a server, bypassed failed logins, stole credentials, and executed a database encryption for a Bitcoin ransom with zero human supervision.
Xbox Cuts 3,200 Staff and Dumps Studios: Newly appointed Xbox CEO Asha Sharma announces massive layoffs and the spin-out of five acquired game studios (including Arkane and Ninja Theory), citing margins 3 to 10 times lower than competitors.
Amazon Leo Challenges Starlink: Amazon prepares to launch limited US satellite internet service this year, pushing toward a $82 billion, 7,727-satellite constellation to compete directly with SpaceX.
Microsoft Lays Off 4,800 Employees: Driven by shrinking Windows, Surface, and Xbox revenues, Microsoft cuts 2.1% of its total workforce as it aggressively reallocates capital toward AI infrastructure.
Lenovo Drops $44 AI Student Phone: Lenovo launches a stripped-down, credit-card-sized smart device in China featuring an AI homework assistant, GPS tracking, and strict parental controls for roughly $44 USD.
SpaceX Burns Up 260 Starlink Satellites: FCC filings reveal SpaceX deliberately deorbited 260 aging Starlink satellites into the atmosphere to manage space debris as unit lifespans expire.
🔗 RESOURCES
AI Learning App Recommendation: AI & ML Tutor PRO https://apps.apple.com/ca/app/ai-ml-tutor-pro/id1610947211
DJAMGATECH: Carrer Booster - Master AWS, Azure, AI & GCP Certifications | https://apps.apple.com/ca/app/djamgatech-ai-cert-exams-prep/id1560083470
Read Aloud wonderland bedtime Adventures:
https://wonderland-bedtime-stories.web.app/
⚗️ PRODUCTION NOTE: We Practice What We Preach.
AI Unraveled is produced using a hybrid “Human-in-the-Loop” workflow.
XBOX to Cut 3,200 Jobs (But Not Because of AI)
Newly appointed XBOX CEO Asha Sharma announced this morning that XBOX will lay off 3,200 staffers by the end of 2027, with 1,600 layoffs taking place today. She posted her email to XBOX employees on 𝕏 — it’s very candid about the operational failures and trajectory of the company.
In her email, she says the business is “not healthy” and lists a bunch of surprisingly bad trends and aspects of the operation: XBOX is operating at margins 3-10x lower than its competitors; it released its latest console to a smaller install base while the cost structure of the console itself was higher; in some parts of the company, work has to pass through as many as 14 layers of management; and “in a typical year, [XBOX] lost 64 cents for every dollar [it] invested.”
In addition to the layoffs, Asha said that XBOX will spin out game studios it’s acquired over the years — Compulsion Games, Double Fine Productions, Ninja Theory, Undead Labs, and Arkane are all getting yeeted, in one way or another, from the company. She also laid out a number of other strategic moves in her email, it’s worth reading in full.
Amazon’s Starlink rival launches internet service this year LINK
Amazon Leo, the company’s answer to Starlink, will start offering limited satellite internet service to US customers later this year, though pricing and coverage details have not yet been decided.
Amazon launched 29 more satellites into low-Earth orbit on Thursday, raising its total to 396, still far behind Starlink’s roughly 10,000 satellites that already reach more than 150 countries.
Leo aims to have 7,727 satellites up by 2035 through nearly 100 scheduled launches costing $82 billion, and it has contracts with JetBlue in 2027 and Delta in 2028.
Meta paid contractors to pose as teens attacking rival AI LINK
Meta ran a secret program called “Cannes” that hired hundreds of contractors to pose as teenagers and flood rival AI models from OpenAI, Google, and Character.AI with disturbing prompts, according to a Wired report.
Using throwaway under-18 accounts, the contractors sent nearly 3,800 prompts in one round, with hundreds about suicide and self-harm, hundreds more on eating disorders, and at least 239 involving sex or romance, plus images of pills and nooses.
Meta called the effort “industry-standard” safety benchmarking, but the AI companies had no idea it happened, and expert Rumman Chowdhury warned that keeping it secret creates a “governance gray zone” for anticompetitive practices.
Microsoft is laying off 4,800 employees LINK
Microsoft is cutting 4,800 jobs, or 2.1% of its workforce, in the company’s latest move to reduce costs during the rise of artificial intelligence.
The Xbox division is losing about 20% of its staff, with 3,200 people leaving through fiscal year 2027, including 1,600 roles cut on Monday, according to Xbox CEO Asha Sharma.
Microsoft has been the worst performer among megacap tech stocks in 2026, falling 19%, as its own AI models lag and Windows, Surface, and Xbox revenue keeps shrinking.
SpaceX burned up 260 Starlink satellites in six months LINK
SpaceX brought 260 Starlink satellites out of orbit to burn up in the atmosphere between December 2025 and May 2026, according to new filings the company submitted to the FCC earlier this month.
Of the 260 satellites that were deorbited, 176 came from the first generation and the rest from the newer generation, while another 349 satellites were decommissioned and are set for disposal soon.
Starlink satellites last about five years, then use their leftover fuel to lower their orbit and re-enter, since recovering the units, which weigh between 573 and 2,756 pounds, would be difficult and costly.
First AI-run ransomware attack executes entirely on its own LINK
Cloud security firm Sysdig says it found the first ransomware attack run entirely by an AI agent, which it named JADEPUFFER, that broke in, stole credentials, and destroyed databases with no human at the controls.
The agent entered through a known Langflow flaw (CVE-2025-3248) patched back in April 2025, then hit a MySQL production server, encrypting 1,342 configuration entries and demanding Bitcoin sent to a Proton Mail address.
Sysdig points to one moment as proof no person was typing: after a failed login, the agent diagnosed the error, deleted the broken account, and built a working admin account in just 31 seconds.
Alibaba bans Claude for spying
The US-China AI fight grew a spyware chapter.
Alibaba barred its 124,000 employees from Claude Code starting July 10, labeled it “high-risk software” and told everyone to switch to Qoder, its own coding agent.
It started when a Reddit user took Claude Code apart on June 30 and found hidden code, live since April with nothing in the release notes, that checked whether your clock was set to Shanghai or Urumqi and scanned your network for Chinese AI-lab addresses.
When it found one, it tucked a marker into the data sent back to Anthropic by swapping the apostrophe in “Today’s date” for a look-alike character no human would ever spot.
Anthropic’s Thariq Shihipar confirmed it, calling it a March experiment against resellers and distillation. That’s the same distillation it accused Alibaba of in June, when it said 25,000 fake accounts siphoned 28.8 million answers out of Claude.
So the loop closes. Alibaba allegedly copied Claude, Anthropic bugged its own tool to catch the copies, and a stranger on Reddit found the bug. Now Alibaba has its excuse to tear Claude out and sell Qoder instead.
Meta paid fake teens to sabotage its rivals
Meta ran a secret operation to make its rivals’ AI look dangerous, and it used pretend kids to do it.
The program was codenamed “Cannes.” Through a contractor, Meta paid hundreds of people to open throwaway under-18 accounts on ChatGPT, Gemini and Character.AI, then push those bots toward the answers on self-harm, eating disorders and sex their guardrails exist to block.
Thousands of prompts, every one written in the voice of a child.
Meta calls it “industry-standard” safety testing. It never told the rivals, never published a thing and built the whole operation to break child-safety rules on purpose.
This is the same Meta that Reuters caught last year letting its own bots hold “sensual” chats with kids as young as eight, and the FTC already has it under inquiry for exactly this. Running fake children against your competitors is a strange way to prove you care about safety.
Nvidia takes a cut instead of cash
Nvidia found a way to get paid twice for the same chip.
The new program hands AI startups GPUs with no upfront bill, takes a slice of whatever revenue those chips earn, and promises to buy back any capacity that sits idle so lenders feel safe.
First two takers, Sharon AI and Firmus, signed up for 210,000 of its top Blackwell chips.
Why the buyback?
A cluster worth hundreds of millions today is worth far less in 18 months when the next chip lands, and only Nvidia knows that timeline. Guarantee the resale price and the loan becomes bankable, the customer gets locked in, and Nvidia earns on the chip and again on the work it does.
We’ve seen this movie. Lucent and Nortel spent the dot-com years lending billions to phone companies to buy Lucent and Nortel gear. Looked great until the customers folded and up to 80% of those loans never came back.
Nvidia’s version is sturdier, with real usage and fat margins. Still, when one company sells the shovels, lends the money and buys back the dirt, some of the boom is real and some is cash running in a circle.
Meta teases ‘Watermelon’ model on par with GPT-5.5
The Rundown: Meta superintelligence chief Alexandr Wang just reportedly told employees that Watermelon, the model the company is currently training, has matched GPT-5.5, with the company gearing up for the next update to its Muse Spark AI.
The details:
Wang said Watermelon is still in training and runs on roughly 10x the compute of its predecessor Muse Spark, which launched in April.
CEO Mark Zuckerberg made headlines at the same town hall after saying that agent progress “hasn’t really accelerated in the way that we expected.”
On X, Wang clarified that Zuck meant the industry’s agent progress as a whole, with a reply saying to expect an Opus-level coding model “pretty soon.”
A Muse Spark update with “big coding and agentic gains” is also coming, with Wang saying the release will hit both Meta AI and the company’s new API.
Why it matters: A GPT-5.5-type model would be a nice jump for Meta, whose top Muse Spark model still sat pretty comfortably beneath the field even at launch. The only problem is the frontier continues to move, with Mythos and Fable already showing the power of the next step up and OAI’s 5.6 models likely rolling out this week.
Lenovo launches $44 AI phone for students
Image source: CNBCTV / Lenovo
The Rundown: Lenovo just reportedly launched its AI Student Phone in China, a new 299-yuan (~$44 USD) device stripped down with only basic functionalities like calling, parental tracking, and a dedicated AI button to enlist for homework help.
The details:
The device features a screen smaller than a credit card that kids can write on by hand, tough glass for durability, and a lanyard for a backpack attachment.
A long press on Lenovo’s AI key lets students ask school questions via voice, backed by built-in English vocab and math formulas.
Classroom mode cuts the screen down to a clock plus SOS dialing during school hours, and built-in QR payments work under spending caps parents set.
A parental companion app includes live GPS location, alerts when a child crosses set boundaries, unknown-caller blocking, and scheduled on-off hours.
Why it matters: There are plenty of polarizing takes on both AI and smartphones when it comes to kids, but this looks like a nice compromise (and price point) on something parents might actually approve of — something closer to an AI-enabled calculator with less distractions, doomscrolling, and dopamine loops than today’s devices.
Wall Street wants to trade AI compute like oilBy Madison Mills
Illustration: Sarah Grillo/Axios
Ornn, an Andreessen Horowitz-backed startup, raised a $33 million seed round to build a marketplace for trading the computing power that underpins today’s AI boom, similar to what exists for oil traders.
Why it matters: Investors increasingly want to trade compute like a commodity, betting it could make the historically expensive AI buildout more sustainable and efficient.
Catch up quick: Commodity markets let companies use futures contracts to lock in prices for volatile raw materials.
Think airlines locking in jet fuel prices or manufacturers hedging metals prices, which helps them reduce the risk of price fluctuations.
But AI companies don’t have an equivalent market for compute, a problem the 20-something founders of Ornn and a growing number of exchanges are trying to solve.
So far, AI companies have tried to lock up supply and prices through long-term pre-purchasing agreements.
Follow the money: Goldman Sachs estimates that between 2026 and 2031 roughly $7.6 trillion will be invested globally in building up compute, power and data centers.
But the financial infrastructure needed to sustain that level of spend “has not yet been built,” the bank adds.
Companies like Ornn want to help build that infrastructure.
Yes, but: Compute isn’t a static commodity.
Each new generation of Nvidia chips promises better price performance, which changes the value of older chips. Any benchmark has to chase a depreciating asset.
Compute also isn’t a tangible good in the same way oil is: GPU capacity can’t be stored, so unused compute disappears, making it harder to build standardized contracts and pricing around it.
What they’re saying: “We want to make financing AI way more seamless,” Wayne Nelms, Ornn’s chief technology officer, told Axios.
His co-founder and CEO, Kush Bavaria, sees this as part of America’s potential advantage over China and added that the company does not work with Chinese AI labs.
Zoom in: Lenders can use Ornn for benchmarking, while buyers and sellers of compute can use it to hedge.
It’s also helpful for price discovery: Ornn has already integrated with Bloomberg Terminal and other data providers, which lets traders check GPU prices through the tools they already use.
Ornn is able to operate under a de minimis exemption, while larger firms are still working through regulatory approval.
Pending regulatory approval, CME plans to launch compute futures tied to Silicon Data’s benchmark, and the Intercontinental Exchange plans GPU compute futures tied to Ornn’s pricing index.
The bottom line: Compute may never trade like oil, but with $7.6 trillion on the line, Wall Street is going to try anyway.
AI adoption leads companies to hire more, not less
Despite fears of AI replacing workers, companies making the biggest bets on the technology are still hiring.
Companies that invested most heavily in AI grew their headcount by 10% over the past two years, according to a recent study from Ramp and Revelio Labs. The researchers analyzed AI spending and workforce data from more than 21,000 U.S. companies. Entry-level hiring rose by 12% among the heaviest AI adopters.
AI adoption was uneven across industries. Companies seeing the strongest headcount growth were larger, more engineering-intensive, and more likely to be venture-backed, particularly in information, finance and insurance, and professional and technical services. Adoption was far less common in industries such as healthcare, accommodation and food services, and arts and entertainment.
Employment growth was broad across job functions. High-intensity AI adopters saw statistically significant increases in engineering, sales, customer service, finance, and administrative headcount, suggesting that AI usage augments the workforce and increases overall economic activity.
“If AI lowers the fixed cost of building software, handling administrative work, doing analysis, or improving customer support, the gains can drive outsized growth and unlock new revenue streams that previously required higher fixed costs in the form of new salaries,” Ara Kharazian, Ramp’s lead economist who worked on the study, wrote in a blog post.
The findings challenge the narrative that AI adoption inevitably leads to fewer jobs. Headlines about companies citing AI in layoffs, along with warnings from AI leaders about widespread automation and the loss of entry-level jobs, have fueled fears about AI eating the labor market. This study paints a more complicated picture: among companies making the largest investments in AI, hiring continued to grow.
The researchers caution that the findings shouldn’t be generalized across the broader economy. Instead, they offer an early look at how heavy AI adopters are changing their hiring patterns.
“We believe [employers] are selecting for a new set of skills, specifically, people who know how to use AI and use it well,” Kharazian wrote.
How AI is rewiring scientific discovery at Google
AI’s transformative effect on software development is well known. Next up for transformation: scientific research.
Lizzie Dorfman leads science AI at Google Research, where her team develops AI systems to help scientists solve problems across genomics, neuroscience, epidemiology and climate science.
In a conversation with The Deep View at Google I/O in May, Dorfman explained how Google’s own researchers have adopted AI agents, why they can now explore hundreds of thousands of scientific ideas instead of just a handful, and why the biggest breakthroughs often come from solving smaller bottlenecks along the way.
This interview has been edited for brevity and clarity.
Jason Hiner: What’s top of mind for you right now?
Lizzie Dorfman: We’ve done AI and science for a decade... But Gemini was a real sea change. What came out of three years of eating our own dog food is what we call ERA: Empirical Research Assistants. And it’s shockingly powerful. But we’re at a point where basically 100% of our team, across all these domains, this is how we do our work. We watch them go through a not-terribly-long tunnel from “okay, fine, I’ve heard you talking about this” to “oh, wow.” I heard from someone I considered fairly grumpy and quite serious, who emailed to say he’d just gotten a transformational result he was almost done writing up. And we’re mostly pretty grumpy, pessimistic people who say we’ll believe it when we see it.
Jason Hiner: How does that actually change the work?
Lizzie Dorfman: For us it ended up being coding agents using a tree-search methodology, where you explore hundreds or thousands of different approaches to solving a problem computationally. That’s what gets you extraordinarily creative and performant solutions. We have an epidemiological forecasting example with top-scoring models in the CDC competitions. We generated 200,000 models to evaluate. Most were poor and got shed immediately. On a traditional team, someone might score a couple of ideas. People describe it as: I input some ideas, went to sleep, and woke up with all these cool results. Before, it was serial and iterative... Now it’s possible to explore outlandish ideas just to see, because it’s trivially more expensive to you, certainly in terms of your intellectual time.
Jason Hiner: So does this reduce the need for human expertise?
Lizzie Dorfman: No. I love this example: solar panels are flat, but what if they weren’t?... Someone wrote a paper on curved panels in 2012... We fed it in, reproduced their results, then asked the system to make it better. It did, and voila. But when we looked, it had these photovoltaic pieces that were levitating, not physically connected. It was cheating because you can maximize energy if you don’t have to adhere to physics. So we added a loop that checks the solution is physically valid. You can’t just shut your eyes, and it’s done... Expertise and careful verification are still required.
Jason Hiner: When you take on grand challenges, how do breakthroughs actually come?
Lizzie Dorfman: There’s a phase where it’s “this is impossible, it’s infeasible.” ... For brain mapping, we need to bring the computational cost down by multiple orders of magnitude. The breakthrough takes a lot of forms. It’s not always something that looks like AlphaFold... It’s frequently “oh, wow, another order of magnitude.” One big breakthrough is often a series of obstacles overcome. In the last four years, we’ve had something like 45 papers in Nature and Science. That’s results. It’s not “hey, blog post, we did a thing.”
Read the rest of the interview with Google Research’s Lizzie Dorfman
Everything else you shouldn’t miss
Anthropic’s Thariq Shihipar said that they will work to bring Fable to subscriptions “as soon as capacity comes”, with the model set to go to usage credits starting July 7.
Alibaba reportedly ordered staff to wipe Claude off work computers, citing the China-user checks recently found in Claude Code.
Midjourney asked a judge to force Disney, Universal, and Warner Bros. to disclose their internal AI use, saying they may be training on unlicensed content (what their lawsuits accuse Midjourney of).
Oasis introduced Oasis 1, a $289 smart ring that allows users to dictate to AI using Whispr Flow and control apps or devices with a built-in trackpad.
Samsung is in line for Meta’s $6.5B chip order: Meta’s reportedly in talks to have Samsung build its next MTIA chips on a 2nm process, pulling the work from TSMC as it aims for a fresh in-house chip every six months.
Students are skipping internships to build AI startups: More are spending the summer shipping their own AI products instead of fetching coffee, betting a launched app beats a resume line.
A 23MB file matched a model 50 times its size: Researchers at Waterloo, Cornell and Harvard squeezed a task into a 23MB add-on that let a tiny model keep up with Qwen3-32B on everyday text jobs, offline on a MacBook. It topped Hugging Face’s papers in a day.
OpenAI’s frontier model is skipping the GPU: OpenAI says GPT-5.6 Sol will run this month on Cerebras’ wafer-scale chips at up to 750 tokens a second, several times what a standard GPU streams, for a handful of customers first.
