- Everyday AI
- Posts
- Ep 661: Out of the Shadows: How to Manage AI Sprawl
Ep 661: Out of the Shadows: How to Manage AI Sprawl
How to fight AI sprawl, Anthropic surprises with Claude Opus 4.5, Meta and Google partner on AI chips, White House's shocking AI announcement and more.
Everyday AI
November 25, 2025
Outsmart The Future
Today in Everyday AI
8 minute read
🎙 Daily Podcast Episode: Shadow AI and AI sprawl are security nightmares. What does it mean? Find out more in today’s show and give it a watch/listen.
🕵️♂️ Fresh Finds: Claude feature updates, ChatGPT packs more punch into Atlas browser, FLUX.2 AI image model drops and more. Read on for Fresh Finds.
🗞 Byte Sized Daily AI News: Anthropic surprises with Claude Opus 4.5, Meta and Google partner on AI chips, White House's shocking AI announcement and more. Read on for Byte Sized News.
🧠 Learn & Leveraging AI: An industry leader at Airia dishes on how to fight back against AI sprawl, which is likely costing your company more than you know. Keep reading for that!
↩️ Don’t miss out: This week's AI news that matters, ChatGPT goes all in on shopping, Google linking gemini to Notebook LM, Amazon's $50 billion AI bet and more. Check it here!
Ep 661: Out of the Shadows: How to Manage AI Sprawl
Even if you've banned AI, your employees are 100% using it. 🥵
To make matters worse?
Even if you've approved a certain AI system, your teams are probably using whatever they want.
And those choices are likely putting your enterprise data at risk.
So, how do you reel in manage the AI sprawl?
Kevin Kiley, the CEO of Airia, is laying out the playbook.
Out of the Shadows: How to Manage AI Sprawl -- An Everyday AI Chat with Jordan Wilson and Kevin Kiley
Also on the pod today:
• Shadow AI risks exposed 💻
• AI spaghetti: real enterprise mess 🍝
• Agent permissions causing vulnerabilities 🚨
It’ll be worth your 31 minutes:
Listen on our site:
Click to listen
Subscribe and listen on your favorite podcast platform
Listen on:
Here’s our favorite AI finds from across the web:
New AI Tool Spotlight – nao is the AI data editor designed for data teams, Klariqo is an AI assistant that handles every phone call, every website visitor, all the time, CatDoes is The fastest way to build production-ready mobile apps. Describe your vision, AI agents will do the rest.
Agentic Browsers — OpenAI released some big updates to its AI browser, Atlas.
AI Wildlife Data — 11,000+ camera-trap videos, 99 species — the largest wildlife video dataset yet
Malaysia AI Funding — Malaysia bets on massive data‑centre build to turn AI funding into global advantage
New Anthropic Agents — On-demand tools, code-driven calls, and real examples — scale agents without context bloat.
AI Images — Black Forest Labs launched FLUX.2, a photoreal image generator with multi-reference, pose control, and clean text at up to 4MP.
Claude Updates — With Opus 4.5, there were some under-the-radar updates from Claude.
AI Chips — Google's chips gain momentum as Meta eyes multi‑billion purchase.
1. Anthropic launches Claude Opus 4.5, claiming a major step for coding and long-running workflows 🧑💻
Anthropic released Claude Opus 4.5 today, saying the model boosts coding, agentic workflows, spreadsheets, and long-horizon tasks while cutting token use and cost compared with prior models.
The company highlights state-of-the-art results on internal engineering and agent benchmarks, improved safety against prompt injection, and new developer features like an effort parameter and expanded platform integrations. Opus 4.5 is available now across Anthropic apps, the Claude API, and major clouds at new pricing intended to make higher-end capabilities more accessible. =.
2. White House launches ‘Genesis Mission’ to turbocharge AI-driven science 🧑🔬
In a timely move, the President signed an executive order creating the Genesis Mission, a federally led effort to build a secure American Science and Security Platform that will pool federal datasets, supercomputers, and AI models to accelerate breakthroughs in areas like energy, biotech, semiconductors, and quantum science. The Department of Energy will run the platform, identify initial data and compute assets, and aim to demonstrate initial capability within about nine months, while the White House coordinates interagency and private-sector partnerships and workforce programs.
The plan emphasizes tight security, standardized partnerships, and clear rules for data, IP, and commercialization, signaling a major federal push to keep the U.S. competitive in AI-enabled research.
3. Meta eyes Google chips, rattling Nvidia stock 📉
According to The Information, Meta is weighing use of Google’s TPUs and may rent them from Google Cloud as soon as 2027, a timely development that sent Nvidia shares down about 4% in premarket trading while Alphabet jumped.
The move would validate Google’s custom AI silicon and mark a meaningful step toward diversifying the dominant GPU-based AI supply chain that Nvidia currently leads. For Nvidia the risk is competitive pressure rather than immediate displacement, because GPUs remain the industry standard while TPUs add another efficient option for large-scale AI workloads.
4. Google rolls Gemini deeper into Workspace with new security and creative AI tools 🧠
Google today expanded Gemini across Workspace apps, pushing timely AI upgrades that add video generation in Vids and Slides, slide creation inside the Gemini app, and stronger data analysis in Sheets.
The update also layers enterprise protections into Drive with AI ransomware detection and bulk restoration, plus client-side end-to-end encryption for Gmail to keep sensitive mail secure even outside the organization. New productivity touches include Gemini in Drive mobile, summarization and tone refinement in Chat, and speech translation in Meet for several European and Latin languages, making AI assistance available where work already happens.
5. OpenAI rolls out “shopping research” in time for holiday shoppers 🛍️
OpenAI announced a new ChatGPT tool called shopping research just as consumers ramp up holiday spending, aiming to deliver detailed, curated buying guides that consider budget and priorities. The feature automatically appears for shopping-related queries, pulls from public retail sites, and may take a couple minutes to generate deeper recommendations while simple queries still use standard ChatGPT replies.
OpenAI says results are organic, will not share chats with retailers, and acknowledges possible errors in pricing and availability, with future plans to integrate Instant Checkout. The rollout covers Free, Go, Plus and Pro users who are logged into ChatGPT, signaling the company’s growing push into e-commerce.
6. Nvidia pushes back on Michael Burry’s “Cisco” warning �*
Nvidia quietly circulated a memo to Wall Street analysts rebutting Michael Burry’s recent claims that today’s AI spending boom echoes the late-1990s telecom overbuild, specifically defending its stock-based compensation accounting, share repurchase math, and customer GPU depreciation timelines.
The company says repurchases were overstated in Burry’s post, that employee equity and buybacks are separate, and that customers legitimately depreciate GPUs over four to six years because older models still hold economic value. Burry argues hyperscalers’ massive capex plans and long depreciation assumptions risk creating oversupply and a Cisco-style collapse, while Nvidia insists its strategic investments are small relative to revenue and that AI startups raise capital mostly from external investors.
Your most productive employee just accidentally sent your product roadmap to China.
Whoopsies.
But… it wasn't a hack.
They were just trying to write a strategy document faster using DeepSeek.
But they didn't read DeepSeek’s terms and conditions.
The fine print explicitly states you have no guarantee of confidentiality and your data is being processed on Chinese servers.
This is exactly why we brought in Kevin Kiley, CEO of Airia, on for today's show.
Kevin is running one of the hottest AI orchestration platforms in the world right now. He sees the data nightmares that make Fortune 500 CISOs wake up in a cold sweat.
On today’s show, we went deep on why "shadow AI" is more dangerous than you think, the specific clauses in free tools that destroy your IP protection, and how to become a "model free agent" before your favorite vendor crashes.
Let’s get it.
1. Stop well-intentioned employees from leaking your data 🇨🇳
You think your security firewall is keeping you safe.
Wroooong shorties.
Your biggest vulnerability is the ambitious team member who thinks your approved tools are too slow.
They want to do a good job.
So they bypass IT and paste sensitive data into free tools like DeepSeek or a free plan ChatGPT just to get the work done.
Here is the nightmare scenario.
Unlike old software that just sat there, these new AI agents have autonomy.
If an employee gives an agent broad permissions to "fix code" and then leaves the company, that agent keeps running.
You basically have a zombie super-user running loose in your system with no one watching the wheel.
Try This:
Run a "Shadow AI Amnesty" hour with your direct reports this week.
Ask them to list every unauthorized AI tool they use for productivity, promising zero punishment for honesty.
Identify the specific capabilities they crave—like coding or writing—and buy the enterprise versions immediately.
It is cheaper to pay for the software than to pay for the data breach.
2. Avoid getting held hostage by one provider 💸
Kevin said that many companies are making a critical strategic error.
They are marrying a single model provider like OpenAI or Anthropic.
Big mistake.
If your business runs on one model and that model goes down, though, your revenue stops.
There is also the cost trap.
The price difference between model generations can be massive—sometimes an 800% swing for similar performance.
You need to act like a "free agent."
Build an orchestration layer that sits between your apps and the models so you can swap the engine under the hood without rebuilding the car.
Try This:
Run your most complex daily prompt through three different models like GPT-5.1, Opus 4.5 and Gemini 3 Pro.
Compare the speed and cost per token rather than just the answer quality.
You will likely find a cheaper model handles the task just as well as your expensive flagship.
Document this winner as your official "failover" default for when your main provider inevitably crashes.
3. Stop building a tangled mess of AI spaghetti 🍝
We have a new term for the hot AI mess most companies are cooking with in the kitchen.
AI Spaghetti.
(Yeah, we told Kevin we’re legit stealing this term.)
AI Spaghetti happens when you throw random AI tools at the wall to see what sticks without a central plan.
Knees week, arms are heavy. And so is the duct tape holding up your laughable excuse for an AI stack.
You have marketing using Jasper, finance using Claude inside of Excel, engineering using Codex (and Curor. And Gemini CLI), and HR using Copilot. And no one knows who’s using what or if it was approved.
What makes this even worse? None of these tools talk to each other.
That’s AI sprawl, at its finest. Err…. worst.
It destroys your ROI because you are paying for redundant capabilities across five different departments while creating massive security holes.
The goal isn't to have the most AI tools.
The goal is to have a "model garden"—a curated list of safe, approved models that employees can use without thinking.
Try This:
Pull your department's credit card statements from the last 90 days.
Highlight every single SaaS charge that includes AI features to see how much redundancy you have.
Pick one official tool for each category and cancel the expensive duplicates by the end of the month.
Take the money you saved and invest it in an orchestration tool that actually secures your data.
Reply