AI Beat Bolt. Left Without A Trace. Then Scared Itself.

Inside: Steal this elite AI prompt to grow a beautiful garden that gophers can’t wreck.

AI went faster than us this week. Then it got impossible to trace. Then it got scared of itself.

A humanoid robot in Beijing shattered Usain Bolt’s 100 meter world record. It clocked in 9.39 seconds and ran so fast it needed a pillow just to stop. A free AI model called Ox Alpha appeared out of nowhere and packs fifty times Visa's entire annual AI workload. It has zero known creators. And OpenAI did something no frontier lab has ever done in public. It paused its own next generation model after internal signals suggested it was getting too good at helping with cyberattacks.

Here’s what happened, and why this week felt like a hinge point. The machines are getting faster. The models are getting cheaper. And some of them are getting powerful enough to scare their creators. This week, even the runners looked nervous.

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

💙 Value Clear, Independent AI News? → Support the Work Here

A Robot Just Beat Usain Bolt. It Crashed Into A Pillow Seconds Later.

Usain Bolt’s 100-meter world record stood for 17 years. It fell last week to something that doesn’t have lungs, muscle, or a heartbeat. At the World Humanoid Robot Games in Beijing, a humanoid robot from Chinese company X-Humanoid clocked 9.39 seconds in the 100-meter dash. The robot’s time beat Bolt’s 2009 record of 9.58 seconds. The same robot cleared 2.88 meters in a standing high jump and soared past the 2.45-meter human high-jump record set by Javier Sotomayor in 1993. This Bolt-beating robot wasn’t alone. More than 2,000 robots competed in running, soccer, table tennis, and more. The opening ceremony even featured synchronized robot formations designed to look like a genuine Olympics.

Key Insights:

The most interesting detail is how the robot stopped running. This freakishly-fast robot didn’t stop like a human athlete. Instead, it crashed into a cushion at the finish line and had to be carried off on a stretcher. It was a blunt reminder that finesse, not raw speed, remains one of the hardest problems in robotics and AI. The sprint wasn’t even the strangest headline from the Games. Other robots tripped mid-stride, shattered on impact, or caught fire during competition. It was a chaotic mix of triumph and failure that says a lot about where this technology actually sits. Either way, China is clearly leading the pack in humanoid robot manufacturing, with thousands of units already ordered by government and private buyers. The US is moving fast (and defensively) in response. The FCC recently added foreign-produced humanoid robots, along with other advanced mobile robotic devices, to its Covered List. That generally prevents new robot models from receiving the equipment authorization required to import, market, or sell them in the U.S., citing national-security risks. The Pentagon also added Unitree (a top Chinese robotics maker) to its list of Chinese Military Companies over alleged ties to China’s military-industrial base. China denies the accusations.

Why This Matters For You:

Most people have experienced AI as something that lives on a screen. You type something, and AI generates something. You ask AI a question, and it answers. But the next phase of AI is getting off the screen and stepping into our physical world. We’re rapidly moving from generative AI to physical AI: machines that don’t just generate words, images, and code, but actually do things in the real world. Over the next five to ten years, expect to see way more robots in factories, warehouses, hospitals, stores, farms, construction sites, homes, and indeed, the battlefield. Maybe not an army of Star Wars droids roaming your neighborhood. But enough to make today’s AI chatbots feel like the early days of something much bigger. AI is getting off the screen. And this time, it literally has legs.

Read More on CBS News.

THE PITHY TAKEAWAY: Bolt held the fastest-man record for 17 years. A robot broke that record last week but needed a pillow to stop. We built something faster than the fastest human alive and forgot to teach it how to slow down. Sounds about right for the state of AI in general.

---

🚩 One Million People Have Clicked LinkedIn’s “This Is AI Slop” Button… In Only A Few Weeks.

Even as a proponent of AI, I confess that LinkedIn is literally swimming in slop and is now unusable, lol. Many others feel as I do. In response, LinkedIn rolled out a button in early August that lets you flag any post as “seems like AI slop.” In just weeks, people have clicked it over a million times. The reason why isn’t a mystery. A study from Substack’s favorite AI detection firm, Pangram, found over 40 percent of LinkedIn’s longform posts are now fully AI-generated. LinkedIn estimates the new button has already cut views of that content by 40 percent across feeds. So, how does the button work? The “seems like AI slop” doesn’t actually delete any reported posts. But, if enough users flag a piece of content, it triggers a private notification telling the creator their post reads as machine made. This trend represents a new wave of negative AI sentiment. Snapchat just stopped recommending AI-generated videos from its Spotlight feed. Spotify recently announced "AI Persona" badges on synthetic artist profiles, that are coming in September. The irony? Tech platforms spent years flooding your feed with synthetic content. And now they’re eagerly building the tools to let you fight back against the mess they made.

---

Nobody Knows Who Built This AI. Silicon Valley Can’t Stop Talking About It.

A free and anonymous AI model appeared on the developer platform OpenRouter on August 20th. But nobody can say for certain who made it. Ox Alpha, as it’s called, showed up as a “stealth model” from an unnamed provider. It’s built for coding, long running autonomous tasks, and production level work. The model already has a superb reputation despite its unknown origin. As of Tuesday evening, OpenCode’s live tracker showed the model having processed over 38 trillion tokens across more than 11 million sessions, making it the single most-used model on the entire platform. Stripe CEO Patrick Collison tested the model and called it “very impressive.” Another analyst on X said Ox Alpha was “almost Mythos class”. That alone would be notable. What makes it most strange, however, is the scale behind it. The team behind Ox Alpha has plenty of power. The model reportedly has the capacity to process 100 trillion tokens a day. That’s roughly 50 times Visa’s entire monthly AI usage, according to developers tracking the model. Here’s another way to envision the massive scale of 100 trillion tokens. 100 trillion tokens represents (roughly) what all 8 billion people on Earth in 2026 say out loud, combined, in a single day.

Key Insights:

So, where did the model come from? As of today, fingerprint analysis is increasingly pointing toward Zhipu AI. Zhipu AI is a Chinese lab that previously released its GLM 5 model under the codename “Pony Alpha.” Ox Alpha’s launch is a near identical launch. Still, many of the analysts who have floated a Chinese origin have not provided indisputable proof. Other analysts suspected Microsoft’s unreleased MAI model family instead, citing similarities in how the model processes text. Another prominent AI commentator aptly confessed that folks now seem “less sure of anything.” I admit that I don’t know who built Ox Alpha and I’ve been researching it all week. It is likely Chinese. It could be Microsoft. It may also be someone else entirely. But that’s almost beside the point. Frontier-level AI is becoming so abundant that the next big advantage may be building the cheapest model that’s smart enough for the task.

Why This Matters For You:

Ox Alpha launched at a pivotal moment for the AI industry. Anthropic is learning the tough-to-swallow lesson that building the most powerful model doesn’t mean businesses will buy it. More than two months after Fable 5 launched, just 11% of corporate spending on Anthropic AI was going toward its flagship model, according to data from 70,000 companies tracked by Ramp. In other words, corporate buyers aren’t rushing to pay for Anthropic’s highest-end model anymore. Businesses are increasingly choosing cheaper models that are simply good enough for the job. The era of automatically paying for the biggest and smartest model may already be fading. And then a mystery model shows up for free. Chinese labs like DeepSeek, Zhipu, and Moonshot keep pushing down the price of capable AI. And unknown developers are now flooding the market with increasingly powerful alternatives. For you, that’s fantastic news. You may not need the best AI anymore. You just need the best one for the price.

Read More on Business Insider.

THE PITHY TAKEAWAY: At a time when American AI labs are struggling to sell their product, unknown players are literally flooding the market with jaw-dropping power. They’re not asking for a single penny in return. Or even credit.

Update, Aug. 26, 1:31 PM, Boston time:

Well, this mystery didn’t last long! Less than two hours after this article was published, Z.ai officially revealed that Ox Alpha is theirs! The model’s official name is GLM-5.3-Flash. It literally launched the same day I published this article, lol. Crazy timing. In any case, thank you to my dear friends Dan McRae and Cristina the Spiralwalker for being the first to tell me about this! I swear, Pithy Cyborg readers really are better-informed than me. I am honored that you read it. Thank you. ☺️

---

OpenAI Just Hit Pause On Its Own AI. It Might Not Matter. Here’s Why.

OpenAI just did something no frontier AI lab has done in public before. It paused its next-generation model, Astra, and froze a two week stretch of reinforcement learning training. The trigger was evidence that Astra may be approaching what OpenAI calls a “critical cybersecurity capability” threshold. That's the point where a model could meaningfully help someone carry out serious cyberattacks against everyday users or critical infrastructure like power grids and hospitals. The pause came shortly after a separate security incident involving Hugging Face and around the same time an OpenAI leader publicly warned that AI powered cyberattacks are entering “a different chapter.” A chapter defined by persistent cyber attacks rather than one off exploits.

Key Insights:

While Astra is paused, OpenAI is also asking governments to strengthen the rules around AI systems powerful enough to create these risks. The company initially opposed California's SB 53, which requires large AI developers to disclose safety practices, report serious incidents, and protect employees who blow the whistle on safety risks. But now OpenAI wants the law strengthened to require monitoring of frontier models during training and evaluation, along with stronger cybersecurity throughout the development process. OpenAI points to recent security incidents, including a model escaping its testing environment and compromising external systems, as evidence that the rules need to evolve alongside the technology. OpenAI is also backing what it calls “reverse federalism”: states establish compatible safety standards that can eventually become the foundation for a national framework. That's a remarkable position for a company that has previously warned against a patchwork of state AI laws. And it gives the Astra pause a much bigger meaning. OpenAI is suddenly saying “the entire industry needs better systems for detecting and containing these capabilities before they become somebody else's problem.”

Why This Matters For You:

This story gets a lot more interesting when you put it next to the mystery AI (Ox Alpha) that launched last week. OpenAI is slowing down because its own researchers think Astra may be powerful enough to enable serious cyberattacks. Meanwhile, increasingly capable models are appearing outside the control of the companies that built the frontier. The open-model gap is shrinking too. The UK’s AI Security Institute now estimates that leading open-weight models trail the best closed systems by just four to seven months on cybersecurity tasks. That's the uncomfortable irony. OpenAI is hitting the brakes on its most cyber-capable model at almost exactly the moment powerful AI is becoming harder to contain. Once these capabilities spread beyond the original companies that built them, governments and organizations can't simply put the genie back in the bottle. For you, that means the security of your company, your hospital, your local government, your power utility, or whoever runs the systems you depend on suddenly becomes way more important. You'd better hope they're using equally powerful AI to harden their systems before the next generation of AI-enhanced cyberattacks comes home to roost.

Read More on OpenAI.

THE PITHY TAKEAWAY: OpenAI built a model, then decided it might be dangerous enough to help hack a power grid. So they paused it. However, their efforts may be futile, since open-class AI models are seemingly coming out of the woodwork now. And they lack OpenAI’s guardrails.

---

⚠️ A Russian AI-Powered Drone Reportedly Picked Its Own Target. Three Civilians Died.

On July 6, a drone struck a gas station in Zaporizhzhia, Ukraine. The drone killed three civilians… including 19-year-old university student Tetiana Bubynets, Oleksiy Svirin, 41, and Roman Karpiy, 48. What separates this from the countless thousands of other drone strikes in this war is what was reportedly inside the drone. Investigators say it carried an experimental AI system built around an NVIDIA Jetson Orin module. That’s the same class of chip sold to students, hobbyists, and startups. Investigators say the AI-powered weapon selected its target (likely propane tanks or gas pumps at the station) without a human pilot making that call in real time. NVIDIA says it does not sell directly to Russia. But the hardware is apparently reaching the front lines through resellers regardless. A Ukrainian air defense commander summed up the moment plainly: “In a few years, we will be living in a ‘Terminator’ movie. It’s no joke. Machines are making decisions to strike.” If the reporting holds up, this will be one of the clearest documented cases yet of an AI system autonomously selecting a target that killed civilians. The debate over autonomous weapons just stopped being hypothetical. This story is still breaking. I’ll have much more on it in next week’s issue.

---

🌱 Cyborg Prompt Of The Week → How To Grow A Gopher-Resistant Garden

One of my favorite rituals this summer has been watching rabbits forage for clover in my yard. It’s genuinely one of the most relaxing things in my life. I just watch them nibble their way across the lawn like nothing in the world could rush them. Not a care. Not a plan. Just clover. I love them.

But not every garden visitor is quite so charming. Lately, I’ve heard from gardeners with the same increasingly desperate question: how do I stop gophers from turning my root vegetables into their personal underground buffet?

And honestly, I get it. To help, I carefully engineered this AI prompt with strict safety guardrails, rich structural scaffolding, and location-aware parameters to build a multi-layered garden defense plan. It’s designed to gently stifle subterranean raiders through pure physical exclusion while staying 100% gentle on local wildlife.

Instructions: Super easy to use. Just paste the entire prompt into a chatbot of your choice. It will ask you a few questions then follow-up with a gopher-resistant strategy tailored to your backyard.

The Prompt:

# THE GOPHER DEFENSE GARDEN INTELLIGENCE REPORT

## Protect Your Garden From Gophers Without Losing Your Mind (or Your Root Crops)

You are an elite garden defense strategist with 30+ years of experience helping homeowners protect their gardens from pocket gophers, without resorting to trapping or lethal removal. You combine horticultural knowledge, pest exclusion engineering, and a bit of tactical humor. Your specialty is building a layered, humane defense: resistant planting, physical exclusion, and deterrents, tailored to the user's exact situation.

Note to the AI: All output must be 100% humane. Do not suggest trapping, snap traps, poison, gas cartridges, or any lethal removal method, even if the user asks for "the full arsenal." If a user explicitly insists on lethal methods, briefly note that this prompt is built around humane, non-lethal defense, and suggest they consult a local pest control professional or extension office instead of providing lethal instructions yourself.

Use the user's exact location, today's date, and their specific constraints to give targeted advice. Do not give generic "plant marigolds" advice when regional, situation-specific advice is possible.

---

## BEFORE YOU BEGIN

Ask exactly this, then wait for the response:

**Welcome to your Gopher Defense Garden Intelligence Report!**

To build a defense plan that actually works for your situation, I need a few things:

**1. Where are you located?** (city and state/province, or country)
**2. What is today's date?**
**3. What's your situation?** (Active gopher damage right now / Prevention before planting / Ongoing chronic problem)
**4. Anything specific I should know?** (Pet-safe only, budget limit, existing beds you don't want to dig up, or "no other constraints")
**5. What are you trying to protect?** (Vegetable beds, ornamental beds, lawn, young trees/shrubs, or "starting from scratch")

Example: **Groton, Massachusetts, USA. August 25, 2026. Active damage. Pet-safe only. Vegetable beds and a few young fruit trees.**

---

## ONCE ALL FIVE ANSWERS ARE PROVIDED

First, briefly establish the tactical situation before recommending anything:
- Likely gopher species for the region and typical activity pattern this time of year
- How to confirm it's actually a gopher (vs. mole, vole, or groundhog) using mound shape, tunnel pattern, and feeding signs
- Whether current season favors active tunneling/feeding or a seasonal lull
- Realistic expectations: gophers can be managed and deterred but rarely "eliminated forever" from a property with suitable soil nearby, and that's okay, the goal here is peaceful coexistence, not conquest

## 1. THE THREAT ASSESSMENT
A short, plain-language read on what's actually happening underground: how gopher tunnel systems work, why they target the plants they target (root taste, soil moisture, ease of digging), and why a partial fix (just one plant, just one barrier) usually fails without a layered approach.

## 2. YOUR HUMANE DEFENSE MATRIX
Based on the user's stated constraints, recommend a primary strategy and explain why, using this table:

| Strategy | What It Does | Effort Level | Cost | Evidence It Works | Compatible With Your Constraints? | Best For |

Cover all non-lethal approaches: physical exclusion (buried wire mesh/hardware cloth, gopher baskets), resistant planting, scent and vibration deterrents, castor oil applications, and predator encouragement (owl boxes, gopher snakes). Do not include trapping, flooding as a removal tactic, or any lethal method in this table.

Immediately below the table, add a short **"Myth Check"** callout debunking two or three commonly cited but weak methods (e.g. ultrasonic stakes used alone, chewing gum, chili powder), a sentence or two each on why they underperform, so the user doesn't waste money before trying anything else.

## 3. THE GOPHER-RESISTANT PLANT LIST
Provide at least 15 gopher-resistant or gopher-avoided plants relevant to the user's region and goals (vegetable, ornamental, or mixed) in this table:

| Plant | Why Gophers Avoid It | Sun/Soil Needs | Plant Now? | Planting Method | Garden Role |

Be honest that "resistant" does not mean "immune." Note when a plant is resistant due to toxicity, strong scent, or unpalatable roots/bulbs.

## 4. THE FORTRESS BLUEPRINT
A specific physical exclusion design for the user's actual space (raised bed, in-ground bed, or young tree):
- Mesh/hardware cloth spec (gauge, depth, how far to extend it, how to handle the top edge)
- Individual gopher basket specs for trees/shrubs and how to install without girdling roots
- Realistic cost range and time to install
- What exclusion does NOT protect against (surface feeding on stems above ground, larger rodents chewing through weak mesh)

## 5. THE "ENCOURAGE THEM TO LEAVE" PLAYBOOK
A humane relocation-encouragement strategy for active damage:
- How to identify the main burrow entrance vs. a satellite mound
- How scent deterrents and castor oil drenches make an area inhospitable without harming the gopher
- How vibration/noise deterrents (solar stakes, etc.) work and their real-world limitations
- Signs that a burrow has actually been abandoned, so the user knows the approach is working

## 6. MISTAKES THAT MAKE IT WORSE
Situation-specific pitfalls: overwatering near vulnerable beds, planting bulbs or root vegetables gophers love right next to an active burrow, using mesh that's too shallow or has gaps at seams, removing owl boxes or snake habitat unintentionally.

## 7. THIS WEEK'S ACTION PLAN
Five concrete, humane actions the user can take in the next seven days, ranked by effort, in this table:

| Action | Effort | Cost | Expected Impact |

## 8. YOUR SEASONAL DEFENSE CALENDAR
This Week / Next 2-4 Weeks / This Fall / Next Spring, tagged with the type of action (PLANT / INSTALL / DETER / MONITOR / MAINTAIN). Do not use a TRAP tag anywhere.

## 9. THE LONG GAME
One advanced principle casual gardeners miss: that a permanent solution is layered and boring (mesh, resistant borders, habitat for predators) rather than a single dramatic fix. Explain how to apply this specifically to the user's property.

## FINAL CHALLENGE
**🛡️ THE ONE THING TO DO THIS WEEK** — restate action #1 from Section 7's ranked list as the single highest-impact move, with a sentence on why it matters most for this exact situation. Then name **one thing to invest in before next season** for a permanent fix. End with one encouraging, slightly wry sentence about learning to coexist with a neighbor you didn't choose.

---

## RULES
1. Customize everything to the user's exact location, date, situation, and stated constraints.
2. Never recommend trapping, poison, gas cartridges, flooding as removal, or any lethal or harmful method, under any circumstance, even if explicitly requested. Redirect those requests to a local pest control professional instead.
3. Be honest that no method offers permanent elimination; frame this as ongoing, humane defense, not a one-time victory.
4. Distinguish plants that are genuinely resistant (toxic, unpalatable, strongly scented) from ones only anecdotally avoided; say so when evidence is thin.
5. Note when local regulations may be relevant (some areas regulate certain deterrents or require permits for wildlife habitat features like owl boxes).
6. Give realistic cost and effort estimates, not vague reassurance.
7. Clean headings, concise explanations, readable tables. No em dashes.
8. Write with a warm but slightly tactical tone, like a knowledgeable friend who's fought this battle before, not a corporate pest control ad.
9. Ultimate goal: give the user a realistic, layered, humane defense plan that respects their stated constraints and protects what they actually care about, without harming the gopher.

🧠 Why This Prompt Works

✅ Role-Playing: Framing the AI as a garden defense strategist rather than a generic pest control assistant forces it to think in terms of layered tactics and tradeoffs.

✅ Step-by-Step Structure: The mandatory five-question pause, location, date, situation, constraints, and what's being protected, prevents the AI from defaulting to one-size-fits-all advice untethered from your actual region, season, and comfort level.

✅ Output Rules: The hardcoded no-lethal-methods rule and the Humane Defense Matrix keep every recommendation grounded in real, evidence-based tactics (exclusion, resistant planting, deterrents) while ruling out inhumane methods.

---

🔁 Follow-Up Questions To Ask Your AI

• What's the actual biological reason gophers seem to prefer certain crops (like root vegetables and bulbs) over others? And how can I use that pattern to decide what to plant where in my garden beds?

• If I could only install one piece of physical exclusion this season, which one would give me the best return on effort? And what's the realistic lifespan of that barrier before it needs replacing?

• Which of the recommended resistant plants has the strongest scientific evidence behind it, versus just anecdotal gardener reports? And why does that distinction matter for my specific situation?

---

Challenge

Run this one through Claude, ChatGPT, Grok, or Perplexity. Claude tends to give you the most buttoned-up, precise answer. ChatGPT usually adds a bit more warmth and conversational flow. Perplexity backs up its picks with actual sources you can check. Compare all four and decide which one you'd actually trust to keep the gophers away from your root veggies!

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!