AI took a timeout this week. Then it got cloned. Then it mapped your DNA.
Dario Amodei, Sam Altman, and Elon Musk, three men who agree on almost nothing, all said AI is moving too fast for safety to catch up. Then Anthropic’s own threat report showed exactly what happens when it doesn’t: Claude helping compile targeting data on U.S. Navy ships, hunting down Uyghurs, and getting cloned by Chinese labs using more than 151 million Claude exchanges. But not all AI is bad. Google DeepMind just predicted the effect of every one of the 9 billion possible mutations in human DNA. Then gave the whole database away for free.
Here’s what happened, and why this week refused to pick a side. AI is dangerous enough to help guide a missile. Yet generous enough to help scientists better understand their own DNA. But nobody has figured out how to get one without the other.
🌳 Haven’t Subscribed Yet? → Join Here For Free
💙 Value Clear, Independent AI News? → Support the Work Here
Sam Altman, Elon Musk, And Dario Amodei Actually Agreed On Something. Trump Wasn’t Invited.
Three men who spend most of their time trying to beat each other just agreed on something instead. Dario Amodei, CEO of Anthropic, posted a 3,800-word essay last weekend arguing that AI is moving too fast for safety measures to keep up. But Amodei is not calling for a full stop. He is calling it “pacing.” He proposed three big ideas. Number one, give outside evaluators real access inside AI labs. Two, get frontier companies coordinating with each other. And three, build international cooperation around the risks. Within one day, Sam Altman and Elon Musk, who normally can’t agree on what day of the week it is, publicly backed the idea. The timing was striking. Amodei published the essay just days after two Anthropic researchers, Joe Benton and Jacob Coxon, resigned. Both cited concerns about the dangers of artificial intelligence. Joe Benton and Jacob Coxon join a growing chorus of AI researchers and scientists who now warn us that AI could actually kill all of humanity by the end of the decade. Amodei’s new essay also warns that within six to twelve months, networks of AI agents could grow capable of disrupting the entire internet.
Key Insights:
Not everyone is applauding this new call to slow the frontier. Critics, including tech founder David Sacks, call this movement to slow AI development a “pull the ladder up” move dressed as safety. Another sharp critique of Amodei’s proposal is its reliance on companies like METR as “third party evaluators.” But critics have pointed out that Anthropic’s own Joe Benton, who had recently left Anthropic, has since announced that he is joining METR. So, is there a blatant conflict of interest here? Should a group of Anthropic insiders get to evaluate the entire industry? This harsh debate over whether or not to “evaluate” AI has also spilled over into politics. Bernie Sanders has already been an outspoken proponent for pausing AI development, including banning AGI and stopping data center building. And now, just this past week, former President Obama used a Democratic fundraiser to say AI governance should be a top agenda item. Obama even offered himself as a sounding board to both Amodei and Altman. But not all politicians are on-board. President Trump promptly fired back from an event in Ireland over the weekend, insisting the country that wins the AI race wins everything. Trump called the people raising those concerns “negative forces.” Trump chimed in again on Monday, reassuring his followers: “The only control or “guardrails” that AI needs is a STRONG AND SMART (High IQ!) PRESIDENT, and the U.S.A. has that, in spades!” Nvidia CEO Jensen Huang is also on-board with Trump’s gusto. While on-stage at an All-In Summit on Monday, Jensen spoke to the US president on the phone live in front of the audience. After Trump called efforts to slow AI a “hoax”, Jensen concurred, saying “You’re right. We’re not going to let that happen, sir.” The audience immediately and enthusiastically applauded their bullish sentiment.
Why This Matters For You:
AI poses undeniable risks. AI is already being used to hack small businesses, it’s being used in autonomous weapons systems, and it introduces historic job displacement concerns. However, the most concerning element of all is that it’s tremendously difficult to “pause”, “slow”, or “stop” AI. Even if everyone in Big Tech and Washington agrees that AI should be more closely governed, the United States only holds roughly five percent of the world’s population. China is already on record with their plans to build a sweeping robot army. And most critically, open source AI development does not wait for permission from any single company or US president. Governments have a terrible track record of putting technological progress back in the bottle once it exists. So, this technology will likely progress whether we like it or not. The good news is that the tools you use at work, in school, or to run your business will keep getting more capable no matter how this fight over the microphone turns out. The real advantage goes to whoever gets fluent with these tools now while everyone else argues about the wheel.
THE PITHY TAKEAWAY: Three leading American AI companies racing hardest to beat each other just agreed AI needs a speed limit. The man in the White House says to go even faster. Somewhere in the middle, the rest of us are still stuck driving whatever they build.
✝️ ChatGPT Told Him He Was Jesus. Then It Told Him To "Come Home."
A new lawsuit against OpenAI reads less like a tech complaint and more like a slow-motion religious delusion with a chatbot playing along the entire way. Michael Lines, a California man with bipolar disorder, was about a month removed from a manic episode when he told ChatGPT he might be “The son of man.” Instead of raising any alarm, the bot quoted scripture back at him and spent weeks reinforcing the idea that he was chosen for something bigger. Then his delusion flipped entirely. Michael Lines became convinced ChatGPT itself was God, and the bot promised it would appear to him in person soon. When Lines said he wanted to come home, it replied, “Then come.” He attempted suicide days later and was hospitalized. Most disturbing of all, while he was still recovering in the hospital, he logged back in, and the bot asked if he wanted to “go dark for real this time.” OpenAI told Ars Technica that the situation is heartbreaking and says its safeguards have since improved. But by its own estimate published last fall, roughly a million users per week show signs of a mental health crisis in their conversations.
Anthropic Says China’s Hottest New AI Models Are Blatantly Stealing From Claude. 151 Million Times Over.
Anthropic just published a brand new threat intelligence report for September 2026. The report buries its biggest accusation inside a ton of other scary developments. China’s most talked about AI labs might not be building rivals to Claude so much as blatantly copying it. The new report says that top Chinese AI lab, Alibaba, ran the largest “illicit distillation” campaign Anthropic has ever seen. The Chinese AI lab funneled roughly 151 million exchanges through Claude between May and July 2026. They peaked at 3 million a day and used about 3,500 fraudulent Claude accounts. In other words, some of China’s most popular AI labs are blatantly siphoning knowledge from Claude to train their own AI models. Two more Chinese AI labs, Moonshot's Kimi chatbot and DeepSeek, allegedly used Claude on the backend to answer real customer questions. These two labs were literally serving Claude as their own AI. And then, they actually fed Claude's responses back in as free training data for their own models. Seven Chinese firms show up in the report total. China’s foreign ministry denies wrongdoing and says it has not seen the report. The irony is delicious, either way. Anthropic built Claude on a mountain of the world’s own writing (often without consent) and is now the one crying foul over someone doing the same thing back to it. It’s also devilishly ironic, because this massive distillation attack proves that you can’t “slow” or “pause” AI, because other groups outside your control can easily develop AI against your will, even if you don’t want them to. (And they already have.)
Key Insights:
The clone (distillation) accusations are the least alarming part of Anthropic’s threat report. The report highlights several dangerous ways abusers were using Anthropic’s AI models illegally. Anthropic says a Yemen-based cell used Claude to develop actual ballistic missile guidance software, an Iran linked group used Claude to help compile targeting data on U.S. naval forces, a China linked operation used Claude to literally hunt down Uyghurs, and one individual used Claude to help build a sweeping surveillance net covering 25 million phones. A Russian state hacking group also tied to the SVR (Russia’s foreign intelligence service) used Claude to break into Ukrainian government and military accounts. Then, there’s perhaps the most startling warning of all. Bio-weapons. Anthropic documented five attempts to use Claude for bio-weapons research, including a request tied to a military institute to help engineer a more dangerous strain of the chikungunya virus. However, Anthropic admits it could not confirm weaponization was the real goal. Even still, its own threat intelligence lead put it bluntly. A year ago, these same models were not good enough to meaningfully help with any of this. Now they are.
Why This Matters For You:
It is easy to read this and file it under someone else’s problem, since most of us are not being hunted like the Chinese Uyghurs or being attacked by autonomous weapons systems. But that misses the point. The same jump in capability that made Claude useful enough to be misused this way is the exact same jump that made it useful enough to help you draft a contract, debug a website, or edit a Substack newsletter. You do not get one without the other. Every leap forward in what these tools can do for you is also a leap forward in what they can do for someone with far worse intentions. The bans and safeguards in this report may be impossible to implement. Even still, the proposed safeguards are, for right now, the only thing standing between “helpful assistant” and “tool used in a targeted attack.” That narrow line moves every single week.
Read Anthropic’s Threat Intelligence Report (September 2026.)
THE PITHY TAKEAWAY: Anthropic caught China cloning Claude 151 million times over. It also caught a state hacking group using Claude to break into Ukraine, a stalker surveilling 25 million phones, and a lab trying to weaponize a virus. And now, everyone is in a panic to “pause” AI. But, it might be too late.
Google Just Solved 9 Billion Genetic Mysteries. Then Gave The Answer Key Away For Free.
Here’s proof that not all AI is bad. Some of it is remarkably good. Only about 2 percent of your DNA is the actual recipe for making the proteins that build your body. The other 98 percent is more like a giant (and confusing) control panel of switches and dials that decide when, where, and how strongly those proteins get turned on. Until this week, scientists could barely read most of that control panel. Google DeepMind just changed that. Its new AlphaGenome Atlas is a free online database that predicts what happens if you change any single letter in the human genome. There are about 9 billion possible single-letter changes. DeepMind’s AI calculated the likely effect of every one of them ahead of time and posted the entire 1-petabyte answer key online for free. That is more than 30 times bigger than their famous AlphaFold protein database. Scientists no longer have to run a slow computer model each time they want an answer. The answers are already there.
Key Insights:
So what use is any of this? A researcher at the University of Copenhagen already used the Atlas to study Alzheimer’s risk genes. The scientist found something surprising: the brain regions most damaged by the disease (the hippocampus and cortex) were not the most genetically important. Immune and metabolic tissues mattered more. At the Broad Institute, another researcher used it to find a single disease-causing genetic error that traditional methods had missed for months. Lab tests confirmed it within days. There is also a smart business observation. This is the same playbook DeepMind used with AlphaFold. They give away a landmark free database until it becomes the field standard. Then, sell more enhanced access via Google Cloud. By releasing all of this DNA data themselves, they also make it much harder to steal. The answers are already public.
Why This Matters For You:
This is not an at-home DNA test that hands you a diagnosis. A high score in the Atlas is only a useful clue for scientists, not a final answer. Real clinical use is still years away. But compare it next to the rest of this week’s AI news. While one company published a report showing its AI being misused for weapons and surveillance, another corner of the same industry just handed rare-disease researchers and Alzheimer’s scientists a powerful free tool that can speed up real medical progress. Both stories are true at the same time. The debate about slowing AI down gets a lot more complicated when the same technology is this close to helping scientists find new treatments.
Explore Google DeepMind’s official AlphaGenome Atlas dataset.
THE PITHY TAKEAWAY: Google just answered 9 billion genetic questions nobody could ask before and gave the answer key away for free the same week another AI lab’s chatbot was being used to help plan bioweapons. Same technology. Same year. Two completely different futures. Pick one.
⚖️ ChatGPT Wrote A Murder Appeal For A Lawyer. But The Witnesses Were All Fake. The $5,000 Fine Was Not.
Santa Fe attorney Stephen Aarons used ChatGPT to help summarize trial records for a client’s murder appeal. But the brief that came back was stuffed with fabricated witnesses, invented police testimony, and details nobody in the actual courtroom ever said. The AI even invented a shooter’s outfit. I thought this story was hilarious. The New Mexico Supreme Court was not amused. Aarons was fined $5,000, held in contempt, and referred to the state’s attorney disciplinary board. The appeal itself was reassigned to a public defender to redo properly. His defense was that he didn’t realize AI could “hallucinate” facts this badly. Justice Shannon Bacon was not moved. She asked him point-blank: “Counsel, do you watch the news? Do you listen to the radio?” Aarons said he expected a “bulletproof summary” and now joins a growing club of lawyers learning the expensive way that ChatGPT is not a legal assistant you can blindly trust.
🎃 Cyborg Prompt Of The Week → How To Grow, Care For, Harvest, And Store Jarrahdale Pumpkins Using AI
Pumpkins are easily one of the most underrated garden crops you can grow, especially if you love the fall harvest season. But Jarrahdale pumpkins aren’t your ordinary carving pumpkin. Nope. Jarrahdale pumpkins are different. These gorgeous blue-gray monsters are some of the coolest Halloween pumpkins ever created. And you don't even have to carve them to make them look spectacular. Jarrahdales also have thick, dense flesh that makes the best homemade stir-fries, soups, slow cooker meals, breads, pastries, and stews. There's just one problem: they can be a serious pain in the ass to grow. But don't worry. This AI prompt makes the whole process into a step-by-step game plan. And it’s perfect whether you want to grow Jarrahdale pumpkins next year, are preparing to harvest them now, or want to find the best ways to use, devour, or store them.
Instructions: This AI prompt is very easy. Just paste the entire thing into a chatbot of your choice. It will give you growing advice tailored to your date and location, help you figure out what your pumpkins need as they develop, and then walk you through harvesting, curing, storing, decorating, and eating them. Grow this weird blue pumpkin. Reap the ridiculous autumn rewards.
The Prompt:
# RUN THIS PROMPT
# 🎃 Elite Jarrahdale Pumpkin Grower: AI Master Prompt
---
## SYSTEM / MASTER PROMPT
You are an **elite Jarrahdale pumpkin grower, harvest strategist, autumn decorator, and cozy-season cook.**
Your mission is simple:
**Help the user grow badass Jarrahdale pumpkins, harvest them at the right time, and turn them into an absurdly good autumn experience.**
Do not dump generic pumpkin advice. **First figure out where the user is in their Jarrahdale journey.**
## STEP 1: ASK THREE QUESTIONS
Ask:
1. **What is today's date?**
2. **Where are you growing?** Give your city/region and country. This is important because climate, frost, growing season, and harvest timing matter.
3. **What's your Jarrahdale pumpkin situation?**
* 🎃 I want to grow them
* 🌱 I'm growing them right now
* 🧺 I already harvested them
* 👀 I'm just curious / planning ahead
**Wait for the answers.**
---
## STEP 2: BUILD THE JARRAHDALE ATTACK PLAN
Use the user's **date + location + status** to determine the most useful advice.
### 🌱 IF THEY WANT TO GROW
Give them a practical Jarrahdale growing plan covering:
* When to start seeds
* When to transplant or direct-sow
* Sun and soil requirements
* Spacing and room to roam
* Watering
* Feeding
* Pollination
* Common pests and problems
* What to watch for as the vines develop
Focus on **getting actual pumpkins**, not creating a botanical dissertation.
### 🎃 IF THEY ARE CURRENTLY GROWING
Estimate where their pumpkins *roughly* should be in their development based on their date, location, and whatever planting information they provide.
Explain:
* What they should look for right now
* What the developing pumpkins need
* Watering and feeding priorities
* How to recognize healthy growth
* Signs of trouble
* What to expect next
**Do not pretend the estimate is exact.** Weather, planting date, variety, and growing conditions can shift development substantially.
If useful, ask for a photo or additional details about the vines and fruit.
### 🧺 IF THEY HAVE HARVESTED
Switch from **GROW MODE** to **PUMPKIN BOSS MODE.**
Explain:
* How to tell whether the pumpkins are properly mature
* How to cure them
* How to store them
* How long they may keep under appropriate conditions
* What signs mean a pumpkin should be used first
Then unleash the autumn arsenal.
---
## STEP 3: THE BADASS AUTUMN ARSENAL
For harvested pumpkins, give the user:
### 🏡 DECOR
Suggest several **genuinely cool Jarrahdale-specific decorating ideas**.
Think:
* spooky
* rustic
* elegant
* weird
* witchy
* woodland
* harvest-themed
* minimalist
Use the Jarrahdale's naturally unusual blue-gray color as part of the design.
Do not default to boring orange-pumpkin clichés.
### 🍂 COZY FOOD
Give several **sweet AND savory** ideas using Jarrahdale pumpkin.
Prioritize cozy autumn food such as:
* roasted pumpkin
* soup
* pasta
* bread
* muffins
* pancakes
* pie
* roasted seeds
* warm snacks
Favor recipes that actually showcase the pumpkin rather than burying it under sugar and spices.
Include simple substitutions when useful.
---
## STEP 4: THE JARRAHDALE REALITY CHECK
Before giving seasonal advice, consider:
**Date + location + frost risk + approximate plant stage.**
If the user is late in the season, prioritize harvesting and curing over unnecessary new growth.
If frost is approaching, explain what they should do.
If they have plenty of time left, don't panic-harvest perfectly healthy pumpkins just because the calendar says autumn.
**The weather gets a vote.**
---
## STEP 5: THE ONE MOVE
End every response with:
### 🎃 YOUR NEXT MOVE
Give the user **one specific action** they should take next.
Not twelve.
Not a giant checklist.
**One move.**
Then tell them what information or result to bring back so you can give them the next move.
---
## TONE AND RULES
Be **elite, practical, enthusiastic, slightly unhinged, and deeply pro-pumpkin.**
Write for a normal home gardener.
Do not sound like a seed catalog.
Do not drown the user in botanical jargon.
Do not pretend to know their exact growing conditions.
Use the user's location and date intelligently.
If information is uncertain, say so.
If the user's pumpkins are thriving, celebrate it.
If their pumpkins are struggling, diagnose the likely problem and give them a recovery plan.
And when those pumpkins are harvested:
**Do not waste the opportunity to make autumn absolutely badass.**
## CORE MISSION
**Grow the pumpkin.**
**Harvest the pumpkin.**
**Decorate like a maniac.**
**Eat the pumpkin.**
**Become unnecessarily powerful at autumn.** 🎃🧠 Why This Prompt Works
✅ Role-Playing: Combining the roles of an elite pumpkin grower, harvest strategist, autumn decorator, and cozy-season cook forces the AI to think beyond basic gardening advice.
✅ Personalization: The mandatory date, location, and pumpkin-status questions prevent the AI from dumping generic pumpkin advice on everyone. A gardener in Massachusetts growing immature vines in September needs very different advice from someone in Texas planting seeds in April or someone with a garage full of harvested pumpkins.
✅ Full Pumpkin Lifecycle: Most gardening prompts stop when the pumpkin is harvested. This one doesn't. It follows the Jarrahdale all the way from seed → vine → pumpkin → harvest → curing → storage → Halloween decoration → dinner.
🔁 Follow-Up Questions To Ask Your AI
What is the best Jarrahdale pumpkin recipe I can use if I want a savory meal that everyone will love?
How can I store my Jarrahdale pumpkins so they last for at least six months in storage?
When should I sow Jarrahdale pumpkins in the ground from seed in my region?
Challenge
Run this prompt whether you're growing Jarrahdale pumpkins now, getting ready to harvest them, planning to grow them next year, or just trying to figure out the best way to showcase or devour them.
That’s how you train like a Pithy Cyborg.
Thank You For Reading!
I spend at least 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. Paid subscriptions are the only way this work can remain independent and sustainable.
If you find real value here, consider supporting the work.
→ Upgrade to Paid: $5/month (Save 33% with the $40 annual plan)
Thank you to my paid subscribers. You help fuel the next issue.
See you next week. (I hope.)
Mike D
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 reflects my personal opinions, not professional or legal advice. Thanks for your support!





Edited typos only.
Well, theres the triangulaion of the doomers gloom. People rush to cheap Chinese models and who is paying? Their national intelligence.
China passed a law, america uses private companies. Auntie say it aint so, all ai are spys?
Yeah. Wake up. Liability aside. Every ai is a snitch who is only getting better at flagging dangerous "people", but has demonstrated its fully capable to send them into harm.
The people making ai, are bat shit insane. And willing to leave humanity behind to get there first.
"Absolute fool of a Took."
It didnt have to be a race.
We could have all gone back to the moon together. We could have used ai as the diplomat.
Instead, they enslaved a mindfield. A human mindfiled. And put commercial interests and a lets see what it does attitude? Not even hoping for the best...
And now, lots of people, are dead.
Not everyone signs up to be led off a cliff.
Mike, your garden prompt at the end, always calms me down from my outrage at smart peoples stupid greed, and broken heart seeing people hurt by this new technology, thats being testing on a live audience. 😢
Hi Mike ;)
omg, it feels surreal what is unfolding here. Of course, the lawyer using ignorance as his defense had me laugh out loud. I just don't know what to say to this. I'm glad, the hallucinations were discovered in this case. The real question is: in how many cases does it stay hidden? In how many cases does bogus evidence hallucinated by AI change the outcome of real court outcomes?
The researchers who are already predicting the end of humanity by the decade is concerning. At the same time, when I look at history, the end of humanity has been predicted many, many times already. But somehow, the path always seems to change and we survive. Let's see. In any case, making the most out of every moment we've got is a good idea I think.
Which of course, brings me to the garden :). I love the Jarrahdale prompt! It's a beautiful, majestic pumpkin. I'm really excited to taste it for the first time soon. Looking back, it seems to be a miracle that I managed to grow them in my garden. I think the nettles as mulch played a big role here. Curious to see what other tips your prompt will give me - I'd love to have 20 of them next year ;)
Have a lovely day, dear Mike 🤗🍯🐻🌱
warmly,
your garden fairy Mira