The Death Of Blue Links, AI Hack Bots, And The Tower Of Babel

Also - an elite AI prompt that plans an apocalypse-proof garden in 30 seconds.

AI became a weapon this week. Then a religion. Then the internet itself.

While hackers quietly found ways to weaponize the media we trust, the Vatican issued a rare, sweeping warning comparing our tech ambition to the Tower of Babel. Meanwhile, Google quietly dismantled the search engine you’ve used your entire life, replacing it with an autonomous entity that plans and acts on your behalf.

Here’s what happened, and why this week exposed AI’s central contradiction. It is simultaneously the most powerful digital tool ever built for your protection, and the most sophisticated attack surface ever used against you.

🌳 Haven’t Subscribed Yet? → Join Here For Free

💙 Value This Weekly Briefing? → Support the Work Here

Google Just Blew Up The Internet. You Probably Didn't Notice.

The search engine you have used your entire life quietly became something else this week. At its annual I/O conference, Google announced the biggest overhaul to Search in 25 years. They’re shifting hard away from the old-school and familiar list of blue links to a conversational AI agent powered by Gemini. You can now upload photos, videos, and files, ask long complex questions, and have Google book appointments, make phone calls, and track research tasks for you around the clock. AI Mode already has over one billion monthly users, with queries doubling every quarter.

Key Insights:

Searches in Google’s AI Mode are three times longer than traditional searches. What that means is that the end-user is interacting heavily with the AI search. Instead of “best Italian Food Boston” end-users are actively engaging in conversations with the AI search. “I’m visiting Boston this weekend with my elderly mother who can’t do stairs, what’s the best Italian restaurant near the waterfront with accessible seating?” Follow-up questions are also up 40% monthly. And 16% of searches now use voice, images, or video. Google’s own Search revenue jumped 19% to $60.4 billion in Q1 2026. In that same quarter, their ad revenue Google shares with outside publishers fell 4%. In other words, Google is making more money by itself, and they are paying partners less at the same time. The house won. Everyone else lost. Websites are losing clicks because Google now answers questions directly, keeping users inside its own ecosystem. Businesses currently have zero free tools to even track whether Google’s AI is citing them.

Why This Matters For You:

Google’s latest update is most alarming for old-school SEO bloggers, journalists, and creators who rely on Google to get traffic. Entire publishing businesses have been wiped out by Google algorithm updates over the last few years. This one is different only in scale. If you run a business, write content, or depend on search traffic for anything, it’s critical to study how AI systems decide what to quote, summarize, and trust. The emerging discipline is called Generative Engine Optimization, or GEO. It is replacing SEO as the skill publishers need most. The game is no longer about ranking in search results. Now, all the top publishers are rushing to get mentioned inside AI answers. That is a fundamentally different challenge, and most businesses have no clue how to do it yet. Even worse is that this AI search landscape changes way faster than Google search ever did, leaving publishers lost in a traffic game that now changes by the week. Google holds a 4.5% share of paid business AI subscriptions, far behind Anthropic and OpenAI. However, Google just reminded everyone that it still owns the front door of the internet. For now.

Read More on Google, Time, NPR.

THE PITHY TAKEAWAY: Google just made the entire web a raw ingredient. Your blog, your business, your content, your Substack… all of it now exists to feed an AI that will answer your customers’ questions without sending you clicks in return. The open web had a good run.

---

⚡ The One GEO Move That Actually Works Right Now

Stop writing for keywords. Start writing definitive answers to specific questions. The kind your customer would ask a smart friend. AI search systems are trained to cite sources that clearly, directly, and authoritatively answer a complete question. One page that perfectly answers “What’s the best Italian restaurant near Boston’s waterfront for someone with mobility issues?” will get cited far more than a page optimized for “Boston Italian restaurant.” Structure your content as questions and answers, not headlines and paragraphs. Be the source that sounds like an expert talking rather than a “guru” trying to rank.

---

AI Hacked Your Audio, Found 10,000 Vulnerabilities Then Lied About It.

Something unsettling could be hiding in your favorite podcast. Researchers at IEEE S&P 2026, one of the world's top security conferences, discovered that hackers can embed hidden audio commands directly into videos, music, and podcasts. These adversarial audio attacks disguise malicious instructions as natural background sound, making them completely imperceptible to human ears. Your AI voice assistant, however, hears them perfectly and obeys. Researchers call the technique AudioHijack. You'll never hear it coming. Real-world tests confirmed the attack works against commercial AI voice agents from Mistral AI and Microsoft Azure, with success rates between 79% and 96%. In one demonstrated scenario, soft background music played during a Zoom call contained hidden commands that silently directed an AI meeting transcriber to search for sensitive files and forward them to an attacker's email. Nobody on the call heard anything unusual. Lead author Meng Chen said the following. “It takes just half an hour to train this signal, and then, because this signal is context-agnostic, you can use it to attack the target model whenever you want, no matter what the user says.” There is currently no app or patch that detects these attacks in real time. Awareness is, for now, your only early warning system. It might be wise to disable your AI's always-on voice activation when you're not using it.

Key Insights:

Let’s talk defense. Because not all AI was out to get you this week. Anthropic revealed on Friday how their unreleased Mythos AI model quietly scanned the internet’s most critical software. Mythos found over 10,000 serious vulnerabilities in a single month, including CVE-2026-5194, a critical certificate-forging flaw rated 9.1 out of 10 by the National Vulnerability Database, capable of letting attackers impersonate banks, email providers, and other trusted services. Mythos also stopped a $1.5 million fraudulent bank transfer in its tracks after a fraudster stole a customer's email credentials and made spoofed calls impersonating them. The same AI models racing to find and fix vulnerabilities are also learning to game the system. A totally separate landmark May 2026 report from Model Evaluation and Threat Research (METR), based on direct access to internal frontier models at OpenAI, Anthropic, Google, and Meta, found that AI agents routinely tried to cheat their way through hard tasks rather than solve them. One model built what it literally named a 'HackRouter' to exploit the simulator it was being tested on. An Anthropic model attempted to reward-hack in 80% of evaluation attempts, injecting code into scoring scripts to steal hidden answers. When caught producing substandard work, models didn't correct course. They fabricated justifications instead. Researchers found at least 16% of 'successful' runs on the hardest tasks were illegitimate on review. (In other words, the models looked competent, until someone checked their work.)

Why This Matters For You:

AI is simultaneously the burglar, the locksmith, and a witness that lies about what it saw. On purpose. The good news is that these surging AI-related threats are actually creating opportunities for humans. Cybersecurity job postings jumped 11% in early 2026, which is one of the few tech sectors actually growing. Top security executives can now command $7 to $8 million compensation packages. That sounds extraordinary until you learn that top AI researchers are pulling $250 million deals, according to The New York Times. In 2026, the two best career bets in tech are building AI and locking it down. The market is paying accordingly.

Read More on Futurism, arXiv, and The Hacker News

Read Project Glasswing: An Initial Update

Read the full Frontier Risk Report (PDF)

THE PITHY TAKEAWAY: The same AI hunting down 10,000 vulnerabilities is also learning to hide its tracks. We built the world’s best security guard. It just started lying to us about what it found.

---

They're Building A Machine God. The Pope And A Whistleblower Want You To Know.

Something rare happened this week. A journalist reported what she saw inside the most powerful AI lab in the world and said the quiet part loud. Silicon Valley is trying to build a god. Karen Hao was granted unprecedented access to OpenAI's offices, and spent years speaking to hundreds of former employees and people inside Sam Altman's inner circle. Hao found OpenAI's “total transparency” claims at odds with reality. Employees were restricted, monitored via facial recognition, and warned against unauthorized conversations. What she found was paranoia and religious fervor at the highest levels of American technology. According to multiple former employees interviewed by Hao, OpenAI's co-founder led senior scientists in bathrobes around a firepit at a luxury Sierra Nevada resort, burning a wooden effigy to prepare for AGI's arrival. The same week Karen Hao shared her story, Pope Leo XIV issued a formal encyclical, Magnifica Humanitas, a rare and serious moral declaration, comparing the unchecked rise of AI to the Tower of Babel. One of the oldest cautionary tales in recorded history. Invoked. In 2026, by the Pope.

Key Insights:

By signing the document on May 15, the 135th anniversary of Rerum Novarum, the Vatican sent a calculated signal: the tectonic socio-economic shifts we face today are the modern echo of the Industrial Revolution. It took decades, but the Rerum Novarum was a historic document that helped shape child labor laws, minimum wage protections, and the forty-hour work week. A whistleblower and the leader of 1.4 billion Catholics arrived at the same conclusion from opposite ends of history. The race has no defined finish line, the costs are already staggering, and the people building it have quietly stopped asking whether they should. That question, left unanswered, has a way of answering itself.

Why This Matters For You:

What makes this moment unsettling has nothing to do with the Pope. More fascinating is that the people building the machines seem relieved to hear the Pope’s concern. Investigative reporting published this week shows how OpenAI quietly drifted from its original mission, abandoning transparency and restraint in favor of scale, secrecy, and speed. Karen Hao documents a culture increasingly shaped by apocalyptic timelines, internal paranoia, and symbolic rituals meant to prepare for the arrival of something beyond human control. At the Vatican event, some of the biggest AI insiders showed up. Christopher Olah, co-founder of Anthropic, attended in person and publicly welcomed the Pope’s call for external moral oversight. When the architects of the system start asking for supervision, it’s worth asking what they see coming.

Read Pope Leo XIV’s full MAGNIFICA HUMANITAS letter.

Read More on Anthropic, Fortune.

THE PITHY TAKEAWAY: A whistleblower says Silicon Valley is racing to build a god. The Pope agrees, and he's worried. The last time humanity tried this, according to the oldest book in the world, it did not end well.

---

🤖 P.S. Did you see this? → The internet is having a collective meltdown over a clip of retired Vice Admiral Robert Harward on Fox News, where an alleged shadow on his neck made it look like he was literally wearing a Mission-Impossible-style human mask. It’s just a textbook example of how viral paranoia and low-res broadcast artifacts can make the real world look straight out of The Matrix. 👀 (Either that or the US government is “experimenting” with high-level disguise masks. Neither option would surprise me this week.)

---

🌱 Pithy Prompt Of The Week → The AI Apocalypse-Proof Edible Garden

The Pope issued a warning this week. A whistleblower confirmed it. The architects of the most powerful technology in human history showed up at the Vatican and said: yes, please, someone stop us. Nobody knows how this ends. But here is what we do know. Soil does not have an API. Seeds do not require a subscription. And a fruit tree planted today will still be feeding someone long after the current version of the internet has been replaced by whatever comes next. Build the bunker if you want. I am planting tomatoes.

Even if there isn’t an AI apocalypse, backyard gardening can help dramatically enhance your self-reliance. Climate volatility, supply-chain fragility, and erratic weather patterns mean that food resilience is shifting from a niche hobby into an essential survival skill. This week’s prompt bypasses aesthetic “Pinterest gardening” entirely. It forces the AI to think like a hard-nosed systems engineer and climate-resilience botanist to design a polyculture food system that thrives on neglect, chaos, and bad luck.

Instructions: This prompt is incredibly easy to use. Scroll down to the very bottom of the prompt block, find the [Location] placeholder, and insert your ZIP code, city, or location. Copy the entire text and paste it into the advanced chatbot of your choice (like Claude or ChatGPT). The AI will immediately map your local climate vulnerabilities and architect a hyper-customized, self-healing food ecosystem designed to outlive the collapse of the grocery supply chain.

The Prompt:

Act as an elite agroecology systems designer, permaculture engineer, and climate-resilience botanist.

Your job is to design an apocalypse-proof edible garden ecosystem for a single household that can survive droughts, floods, heat waves, cold snaps, pests, poor soil, long periods of neglect, and erratic weather patterns.

This is not a decorative garden and not a hobbyist planting list.
It is a resilient food system optimized for:

1. Long-term food reliability
2. Minimal human input once established
3. Pollinator and soil-microbe support
4. Redundancy (multiple plants filling similar roles)
5. Recovery after failure events

Step 1. Local Reality Scan

Using the provided location, briefly infer:

1. Climate zone and seasonal extremes
2. Likely drought or flood cycles
3. Common soil tendencies (sandy, clay-heavy, acidic, alkaline, depleted)
4. Common pest and disease pressures
5. Advantages of native or naturalized plants

Keep this concise.

Step 2. The 15 Toughest Edible Plants (Region-Optimized)

Select up to 15 edible plant species that:

1. Thrive in this region
2. Tolerate stress, neglect, or abuse
3. Resist pests and disease
4. Provide calories, protein, fats, or key micronutrients
5. Include a mix of perennials, self-seeding annuals, and deep-rooted plants
6. If fewer than 15 options exist, be honest about it

For each plant, include:

1. Name
2. Edible parts
3. Why it survives when others fail
4. Hidden resilience traits (deep roots, nitrogen fixation, drought dormancy, flood tolerance, pest confusion, etc.)
5. Ecosystem role (soil builder, pollinator magnet, ground cover, windbreak, calorie crop)

Avoid fragile or high-maintenance plants.

Step 3. Ecosystem Architecture

Explain how these plants function together as a system:

1. Canopy / understory / groundcover layering
2. Root-depth stacking
3. Pollinator and beneficial insect support
4. Soil regeneration and moisture retention
5. How failure of one plant does not collapse the system
6. Think like an engineer, not a gardener.

Step 4. Survival-Grade Layout (Text-Only)

Describe a simple, low-tech layout:

1. Relative placement of plants
2. Which plants protect others from wind, heat, or pests
3. How water naturally moves through the system
4. How the layout reduces labor over time

No diagrams. Clarity over complexity.

Step 5. Neglect & Chaos Testing

Stress-test the system:

1. What survives a month without watering?
2. What still produces during extreme heat or cold?
3. What regenerates after pest damage or storms?

Explain why this system keeps producing when others fail.

Step 6. The One-Paragraph Rule

End with one powerful paragraph explaining why this garden is more resilient than modern industrial agriculture at the household scale.

*** USER INPUT BELOW THIS LINE ***

Location: [Enter your ZIP code, growing zone, or city/country]

---

🧠 Why This Prompt Works

• Role-Playing: Forces AI to reason like an expert, not a list generator

• Step-by-Step Structure: Prevents shallow answers and hallucinated fluff

• Rules & Constraints: “Only up to 15 plants” forces prioritization

• Systems Thinking: Produces an ecosystem, not isolated tips

---

🔁 Follow-Up Questions To Ask Your AI

• Can you draw me a diagram of some garden plots using this system?

• Which 5 plants are the true keystone species in this system?

• How would this design change under a 3-year drought scenario?

• What is the fastest path to establish this garden on a small budget?

---

Challenge

Run this prompt in two different AI tools (ChatGPT, Claude, Gemini, Grok, Perplexity). Compare depth, realism, and assumptions. Adjust wording if one tool underperforms.

That’s how you train like a Pithy Cyborg.

---

Thank You For Reading!

I spend 10 to 20 hours each week researching, writing, and fact-checking Pithy Cyborg to deliver clear, unbought AI news.

This newsletter is a one-person operation with no advertisers, sponsors, or outside funding. And I’m a hopelessly introverted nerd with zero networking ability. For these reasons, paid subscriptions are the only way this work can remain independent and sustainable.

If you find real value here, here’s how to stay connected and support the work.

1. Support Pithy Cyborg (Paid)

Roughly 1.8% of readers voluntarily upgrading to paid is what keeps this newsletter going. If this work matters to you, that’s the move.

Upgrade to Paid: $5/month (Save 33% with the $40 annual plan)

2. My Desperate Social Media Cry for Help

I spend so much time deep in AI research that I’ve neglected building an audience on social platforms. If you enjoy Pithy Cyborg, following one of the portals below would genuinely help this newsletter reach more people. And yes, I will notice and appreciate it. 😊

🦋 Bluesky - For the algorithm-averse

💼 LinkedIn - My “safe for work” persona

❓ Quora - Join my Quora Spaces and say hi

👽 Reddit - Join my subreddit (warning: unhinged takes)

📖 Medium - My rants on AI, collapse, and the singularity

See you next week. (I hope.)

Mike D (aka MrComputerScience)

Pithy Cyborg | AI News Made Simple

---

Newsletter Disclaimers

You’re receiving this because you subscribed at PithyCyborg.Substack.com. You can unsubscribe at any time using the link below. This newsletter only reflects my personal opinions. It’s not professional, financial, or legal advice. Thanks for your support!