AI DUNN Right Weekly — Issue #29
- Apr 20
- 9 min read
Practical AI insights for business growth
Hey AI Innovators!
This was a week that felt like something shifted. Anthropic built an AI model so powerful they decided the world isn't ready for it. Meta is officially back in the AI race. And Sam Altman is telling the government to start preparing for a world where AI does most of the work. No big deal, right?
If you ever feel like AI news moves too fast... this week was a lot. Let me break it all down for you.
Here's what you need to know:
Anthropic built an AI too dangerous to release publicly — and it found a 27-year-old bug in minutes
Meta's new AI lab finally shipped its first model, and it's actually competitive
Sam Altman wants to tax robots and give every American a dividend check
Perplexity now connects to your bank account
Google quietly dropped a free AI dictation app that works completely offline
Read time: 7 minutes
Big Story
Anthropic Built an AI So Powerful It's Keeping It Locked Away
Here's a sentence I didn't expect to write this year: one of the world's leading AI companies just announced a model and then immediately said you cannot have it.
Anthropic unveiled Claude Mythos Preview this week, and the story behind it is genuinely wild. This isn't a new chatbot or an upgraded assistant. Mythos is a cybersecurity-focused AI that can autonomously hunt for the kinds of software vulnerabilities that human researchers spend months trying to find. In testing, it flagged a bug in OpenBSD that had been hiding for 27 years. Twenty-seven years. It also found a flaw in popular video software that automated scanning tools had missed after five million attempts.
That's not a typo. Five million scans. Missed. Mythos found it.
Anthropic isn't making Mythos publicly available. Instead, the company launched something called Project Glasswing, a coalition of 40+ organizations including Apple, Amazon, Google, Microsoft, Nvidia, and CrowdStrike. These partners will use Mythos to find and patch critical vulnerabilities in major systems before bad actors can exploit the same capabilities. Anthropic is putting up $100 million in Claude credits to fund the effort.
The name is intentional. A glasswing butterfly hides in plain sight using transparent wings. Critical software bugs have done the same thing for decades, buried in complex systems that humans couldn't fully audit. AI just changed that.
And it's not just the cybersecurity angle that's significant. Anthropic's revenue has tripled this year to over $30 billion. Enterprise demand for Claude is so strong the company just locked in a 3.5 gigawatt compute deal with Google and Broadcom to keep up. Anthropic's own chief science officer said Mythos is "the least capable model we'll have access to in the future."
Let that sink in for a second.
An AI that finds 27-year-old bugs in minutes is the least capable version of what's coming.
My take on this?
This is the moment I point to when people ask me if AI is "just hype." Anthropic made a product decision to not release a model because it's too powerful. That is not a PR move. That is a company taking seriously the idea that AI capabilities can cause real harm if they get into the wrong hands first. For business owners, the takeaway isn't panic. It's this: the security landscape just changed, and the tools defending your systems are getting dramatically better, fast.
What's New This Week
Meta Is Back in the AI Game with Muse Spark
Remember last year when Meta's Llama 4 landed with a thud? This week was different. Meta's newly formed Superintelligence Labs, led by Scale AI founder Alexandr Wang, shipped its first model: Muse Spark.
Muse Spark handles text, voice, and images. It has a "contemplating mode" where multiple AI agents debate each other on hard problems. Benchmark-wise, it sits competitively alongside Claude and GPT on reasoning tasks, though it's still catching up on coding.
Where it genuinely leads is health reasoning, which makes sense given Meta's push toward personal AI that knows you deeply. The model is rolling out across WhatsApp, Instagram, Facebook, Messenger, and even Ray-Ban AI glasses.
Is it the best model in the world right now? No. But Meta is back at the table with real resources, 3 billion+ daily users, and a team that rebuilt their entire AI stack from scratch. Don't count them out.
Sam Altman Wants to Tax Robots and Give You a Check
OpenAI published a 13-page policy document this week titled "Industrial Policy for the Intelligence Age" and it reads like... a lot. Sam Altman is asking the U.S. government to tax AI-driven profits, create a national wealth fund that pays dividends to every American (think Alaska's oil fund), implement a 4-day workweek, and build "containment playbooks" for AI systems that can't be shut off.
Axios called it "the most detailed blueprint any tech titan has ever published for how to tax, regulate, and redistribute wealth from the technology he's building." Which is either inspiring or terrifying depending on your worldview. What I find interesting is that the CEO of an $852 billion company is doing this at all. You don't write that document unless you genuinely believe the disruption is coming. For small business owners, the message between the lines is: start thinking now about how AI changes your workforce, your costs, and your competitive landscape.
Perplexity Now Connects to Your Bank Account
Perplexity rolled out a Plaid integration this week, which means you can now connect your checking accounts, credit cards, loans, and brokerage accounts directly to its Computer agent. Ask it to build you a budget, track your net worth, create a debt payoff plan, or prep a retirement dashboard, all through plain-text conversation.
This is a huge pivot. Perplexity started as a search engine trying to out-Google Google. Now it's going after Mint, TurboTax, and your personal finance apps all at once. The company's revenue hit $450 million annually in March, a 50% jump in a single month. The agentic strategy is clearly working. If you use Perplexity already, this feature alone might be worth exploring for a monthly financial check-in.
Anthropic Just Made AI Agents Smarter and Cheaper at the Same Time
Anthropic launched something called the Advisor Tool in beta this week, and it's clever. The idea is simple: instead of running expensive Opus-level AI for every task, you pair a cheaper model like Sonnet or Haiku as the "executor" and only call in Opus as an "advisor" when the task gets genuinely complex. One line change in your API call.
The numbers are pretty striking. Haiku paired with Opus advisor more than doubled its performance score on a complex browsing task, while still costing 85% less per task than running Sonnet alone. For business owners building AI workflows or using platforms that charge by usage, this kind of efficiency architecture means you can get better results without a bigger bill. Keep an eye on this one as platforms start adopting it.
Tool of the Week
Google AI Edge Eloquent
Google quietly dropped a free AI dictation app this week called AI Edge Eloquent, and I think it's going to become a daily habit for a lot of business owners. Here's the thing that makes it different from every other voice-to-text tool out there: once you download the speech model, it runs entirely on your device. No internet required. No audio sent to the cloud. Just you talking, and clean polished text appearing.
Once you speak, it removes filler words like "um" and "uh," and it restructures your rambling into actual sentences. You can choose output styles including Key Points, Formal, Short, and Long. It also imports custom names and jargon from Gmail so it knows your vocabulary.
Who is this for? Honestly, anyone who thinks faster than they type. If you've got ideas that come to you in the car, on a walk, or mid-meeting, Eloquent captures them cleanly without you needing to clean up a transcript afterward. It also saves all your sessions and makes them searchable.
Right now it's iOS only in the U.S., but Android support appears to be coming based on the App Store listing. It's free. There's no subscription. Try it this week if you have an iPhone and you dictate anything at all for work.
Quick Hits Worth Your Time
→ Anthropic's revenue tripled to $30B. While fighting Pentagon scrutiny and rate limit headaches, Anthropic's enterprise customer base doubled and the company locked in a 3.5 gigawatt compute deal with Google and Broadcom to keep up with demand. Claude is clearly not slowing down.
→ HeyGen released Avatar V, and AI video just leveled up. From a 15-second phone recording, Avatar V builds a full video avatar that captures your real facial details, movements, and gestures. Record once, then swap outfits and backgrounds without filming again. The company says it outperformed Google's Veo 3.1 in blind tests. This one is worth watching for anyone doing video content.
→ OpenAI hit pause on Stargate UK. The massive UK data center project was supposed to launch in northeast England, but OpenAI cited high energy costs and unclear regulation as reasons to hold off. The UK government says it's still working with AI companies, but this is a setback for Britain's plans to be a major AI infrastructure hub.
→ Oxford researchers trained an AI to catch heart failure five years early. The system reads fat around the heart from routine CT scans, spots invisible inflammatory signals, and flags high-risk patients with 86% accuracy across 72,000 patients. One in four patients in the highest-risk bucket developed heart failure within five years. Oxford is already working with regulators to bring it to hospitals. This is what AI in healthcare actually looks like.
→ The New Yorker ran a deep investigation into Sam Altman. Drawing on 100+ interviews, internal memos, and private notes from Anthropic's Dario Amodei and former OpenAI chief scientist Ilya Sutskever, the piece alleges a long-running pattern of deception at the top of OpenAI. No smoking gun, but a detailed and concerning picture. Worth a read if you follow the AI industry closely.
Prompt of the Week
With AI agents doing more and more for businesses, one of the most useful things you can build is a weekly intelligence brief about your own business. This prompt helps you do exactly that.
Prompt of the Week
You are my Weekly Business Intelligence Analyst. Using the information I provide below, create a concise Weekly Business Brief I can review every Monday morning.
My business context:
- Business type: [e.g., copywriting agency / retail shop / consulting firm]
- My top 3 priorities this month: [list them]
- My biggest current challenge: [describe it]
- Key metrics I track: [e.g., revenue, leads, client count, hours billed]
Here is my update from this past week:
[Paste in notes, emails, or key events from your week — even rough bullet points are fine]
Please create:
1. A 3-bullet summary of what happened last week and what it means
2. A "wins and risks" section — one or two of each
3. The single most important thing I should focus on this week, with a brief reason why
4. One question I should be asking myself that I'm probably not
Keep it tight. No fluff. Write it like a smart advisor who respects my time.
Why this works: Most of us are too close to our own businesses to see patterns clearly. Feeding your week's notes into this prompt forces a structured outside-in view of what's actually happening. The "question I should be asking" piece is where the real value shows up, because AI will often flag something obvious you've been ignoring.
My Take
This week felt like one of those weeks where you look up from your screen and think... okay, this is actually moving fast now.
An AI that finds 27-year-old security bugs in minutes. A social media giant rebuilding its entire AI stack and shipping a competitive model in under a year. The CEO of OpenAI writing a 13-page document about taxing robots and wealth redistribution. All in the same week.
I know that can feel overwhelming. I get it. There's a lot coming at us. But here's what I keep coming back to: the businesses that are going to thrive in the next few years aren't the ones that have the most AI tools. They're the ones that know how to ask the right questions with the tools they already have.
This newsletter exists because I believe you don't need to be a tech insider to understand what's happening. You just need someone to translate it. That's what I'm here for, every single Monday.
The Anthropic Mythos story hit me hard this week. Not because it's scary, but because it's a reminder that the AI capabilities available to everyday businesses are still miles behind what's being built behind closed doors. Which means there is so much runway ahead. So much to learn, to build, and to take advantage of.
Keep showing up. Keep experimenting.
You're already ahead of the curve just by being here.
See you next week.
Jackie
Comments