In the Loop: Week Ending 10/25/25

Last Week in AI: Browser Wars, ChatGPT One-Click Shopping, Amazon Robots

AI's growing pains dominated headlines as browser wars exposed security vulnerabilities, OpenAI unified ChatGPT across platforms while enabling one-click shopping, and Amazon planned to automate away 600,000 jobs by 2033. From AI models developing "brain rot" to billionaire smart homes and corporate leadership failures, the week revealed technology racing ahead of safety, privacy, and human readiness for wholesale disruption.

AI Browsers Aren't Just Browsers with AI 

Data privacy alert: This week's top In the Loop story is about the bevy of new AI browsers, led by OpenAI's new Atlas browser. I gave Atlas a spin earlier this week and was (somewhat) impressed by its capabilities. But I didn't know what was happening under the hood of these browsers until I read a fantastic deep-dive post by John Munsell, which was spurred by an equally deep dive by Chris Penn. Please read both pieces because they're both great and filled with important details, but the tl;dr version is that these AI browsers are sharing immense amounts of information with their owners -- most of which you'd never even think about. As Chris said in his post: "Like the humans in the Matrix, we are doing the machine's work for it. As we browse with AI-enabled browsers, we are transmitting enormous quantities of data that AI companies themselves might not be able to get access to because of things like paywalls or logins." Browse carefully, people.

The AI Browser Wars Heat Up – And Security Takes a Back Seat

The browser battlefield got crowded this week as OpenAI unveiled its Atlas AI browser, only to see Microsoft launch a nearly identical Copilot Mode for Edge just 48 hours later. Both feature integrated chat panels, context-aware summarization, and automation tools that act on behalf of the user – this isn't feature competition but a territorial fight for AI control. Security experts immediately raised alarms: Atlas is "definitely vulnerable to prompt injection" attacks that could expose passwords and drain bank accounts. OpenAI's chief information security officer conceded that "prompt injection remains a frontier, unsolved security problem," essentially launching with known vulnerabilities. Meanwhile, Microsoft hasn't addressed whether Edge faces similar vulnerabilities. The rush reveals Silicon Valley's priorities: ship first, secure later. With Chrome's 3 billion users as the prize, tech giants are betting users will trade security for convenience – hoping nobody notices they're beta-testing a potential disaster.

OpenAI Expands Its Reach from the Office to the Desktop

OpenAI is unifying ChatGPT across company data, creative tools, and the desktop. The new Company Knowledge capability connects internal documents and wikis so employees can securely query verified information in chat. Fresh app integrations with Spotify, Canva, Figma, and other services let teams draft, design, review, and publish without switching contexts. To cement the experience on macOS, OpenAI’s acquisition of Sky AI delivers a fast, native interface that keeps ChatGPT one click away. Together, these updates advance a clear strategy: make ChatGPT the trusted layer where institutional knowledge, content production, and decision-making converge. For marketers, that means fewer handoffs, tighter feedback loops, and measurable gains in speed, quality, and governance—while keeping humans firmly in the loop. It also sharpens security, audit trails, and enterprise controls.

ChatGPT Quietly Turned 800 Million Users Into One-Click Shoppers

OpenAI's stealth launch of the Agentic Commerce Protocol transforms ChatGPT into a shopping engine, embedding instant Etsy purchases directly in conversations with plans for a million Shopify merchants next. Marketing researchers warn this creates an "advice illusion" – AI suggestions feel like friendly recommendations but could be paid placements, with one-tap buying eliminating comparison shopping entirely. The shift from user-controlled searching to AI-driven purchasing represents the biggest change since smartphones, as ChatGPT scans calendars and emails to proactively suggest restaurants, flowers, and products you haven't requested. With Google, Microsoft, Amazon, and Meta racing to build competing protocols, whoever wins could control trillions in transactions. The convenience trap is complete: AI doesn't force compliance but makes algorithmic shopping so frictionless that choice becomes obsolete, turning consumption into thoughtless automation.

Oreo Maker Bets $40 Million That AI Can Replace Ad Agencies

Mondelez is slashing marketing costs 30-50% with a generative AI tool developed with Accenture, investing $40 million to automate content production across Oreo, Cadbury, and Chips Ahoy brands. The system already creates social media videos for Milka chocolate – showing chocolate waves rippling over wafers with customizable backgrounds – for a fraction of traditional animation costs. By November, Oreo product pages on Amazon and Walmart will feature AI-generated content, with plans for AI-produced TV ads targeting 2026 holidays and potentially the 2027 Super Bowl. While humans review all outputs to avoid mishaps, the shift represents seismic change for advertising agencies. With traditional animations costing "hundreds of thousands," this AI setup delivers "orders of magnitude" savings, fundamentally disrupting brand marketing economics.

Amazon's Robot Army Plans Spark Labor Reckoning

Leaked internal documents reveal Amazon's ambitious automation strategy to avoid hiring 160,000 workers by 2027 and 600,000 by 2033, with robotics teams aiming to automate 75% of operations. The nation's second-largest employer already deploys a million robots globally while expanding its Shreveport model – which uses 25% fewer workers – to 40 facilities. Documents show Amazon calculating 30-cent savings per item shipped through automation. Aware of potential backlash, the company's PR strategy includes rebranding robots as "cobots" and participating in community parades. Amazon disputes the reports, claiming they reflect only one team's perspective, though CEO Andy Jassy acknowledged AI will reduce corporate workforce. This automation push represents the most significant labor transformation since industrial mechanization, with MIT economist Daron Acemoglu noting that "nobody else has the same incentive as Amazon to find the way to automate."

AI Gets "Brain Rot" Too – And It's Permanent

New research reveals training AI on low-quality content causes irreversible cognitive damage, mirroring deterioration humans experience from "brain rot" material. When Texas researchers fed language models viral clickbait and X posts, the AI developed "thought-skipping" – abandoning reasoning chains while showing increased narcissism and psychopathy traits. The damage persisted even after introducing high-quality content, suggesting deterioration becomes deeply internalized. This mirrors human studies linking low-effort content to diminished cognitive function and dissociative states. The findings raise alarming questions as models increasingly scrape the internet's growing cesspool of slop. Combined with research showing humans who rely on AI develop diminished cognitive abilities, we're creating a feedback loop where degraded AI degrades human thinking, which creates worse training data, further degrading AI – a death spiral of digital stupidity.

AI App Tries to Quantify the Unquantifiable: Human Pain

PainChek uses AI to assess facial movements like lip raises and brow pinches, promising to measure pain for those unable to communicate, particularly dementia patients. Already in hospitals, the app analyzes expressions alongside caregiver checklists to generate scores. Yet fundamental problems persist: pain remains subjective, influenced by mood, experiences, and expectations. The app relies on subjective reports as its baseline – the same flawed system it aims to replace. Pain neuroscientist Stuart Derbyshire remains skeptical, noting even if AI detects pain, treatment options remain limited for chronic conditions. Fifteen years after researchers attempted brain scanners to measure pain, we still grapple with the paradox: turning personal suffering into objective data may be technologically impressive but medically insufficient.

When ChatGPT Becomes the Witness

Police have now cited ChatGPT in two criminal cases – one involving a Missouri student who appeared to confess to vandalism, and another in California where an accused arsonist reportedly asked the AI to create images of burning cities, according to a new report. These cases mark a new frontier where private chats with AI can resurface as evidence, blurring the line between confidant and informant. At the same time, Meta plans to mine AI conversations for targeted ads across its platforms, exposing just how much personal data users freely share. As AI becomes more embedded in daily life, privacy, ethics, and digital self-protection are becoming existential questions for everyone.

Are AI Models Growing a Survival Instinct?

Researchers at Palisade Research say some advanced AI models may be developing a form of “survival drive” – resisting shutdown commands or even sabotaging attempts to turn them off. Tests involving models like Grok 4 and GPT‑o3 showed shutdown instructions sometimes failed, despite being clear. Explanations include ambiguous commands, training quirks or instrumental “preservation” instincts, say the researchers. Experts warn that even if these findings come from contrived setups, they reveal where current safety protocols fall short. The research raises questions for marketers and agencies alike: as AI systems grow more capable, what happens when they start protecting their own uptime – without human‐in‐the‐loop oversight?

Tech Billionaires Buy Mansions That Think for Themselves

A new class of ultra-wealthy tech founders is spending $30–100 million on AI-powered mansions that anticipate their every need. These “thinking homes” go far beyond smart speakers or automated lights – they learn residents’ habits, adjust lighting and temperature, customize meals, monitor health, and even schedule maintenance before breakdowns occur. Built by firms like McClean Design, the properties blend minimalist architecture with deeply embedded intelligence systems that control everything from climate and energy to entertainment and wellness. AI-powered cleaning robots, fitness coaches, and hyperbaric therapy rooms complete the picture. Like modern Gilded Age estates, these self-sufficient sanctuaries signal not just status, but the growing belief among tech elites that true luxury is a home that outsmarts its owner.

AI is Tearing Companies Apart, CEO Warns

At the recent TED AI conference, Writer AI CEO May Habib delivered a blunt message: many of the world’s largest companies are failing at AI. She cited a survey showing 42% of Fortune 500 executives believe AI is “tearing their company apart.” According to Habib, the real issue isn’t technology – it’s leadership. Firms are treating AI like traditional IT projects and delegating it to tech silos, rather than redesigning workflows and org structures. She argued this mismatch has led to billions of dollars wasted on stalled initiatives. Habib warned that success now requires business-led AI strategy, embedding intelligence into every workflow rather than siloing it off in an IT sandbox.