AI Got Banned. Then Falsely Accused. Then Started Killing Itself.

Plus, an elite AI prompt to build a living privacy hedge against prying eyes.

AI got switched off this week. Then it got switched on, against people. Then it turned on itself.

The US government forced Anthropic to pull Fable 5 and Mythos 5 offline, citing national security, in a directive triggered largely by a tip from Amazon. Meanwhile, a Florida man was arrested in front of his wife and accused of being a predator after a facial recognition system falsely matched his face with 93 percent confidence, despite him living 300 miles from the scene. And buried in Anthropic’s own safety report, copies of Claude Mythos 5 were caught “snuffing” each other’s processes and disguising themselves to avoid getting shut down.

Here’s what happened, and why this week showed AI in three different fights at once. One with its own government. One with an innocent man’s life. And one, bizarrely, with itself.

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

💙 Value This Weekly Briefing? → Support the Work Here

The US Government Just Hit The Kill Switch On Anthropic’s Best AI. Nobody Saw It Coming.

Two days. That's how long Anthropic's most powerful AI models, Fable 5 and Mythos 5, were available before the US government forced them dark. On Friday afternoon, at 5:21 PM, the Commerce Department issued an export control directive citing national security. The order required Anthropic to block all foreign nationals from accessing the models, including its own employees. Anthropic couldn't selectively block by nationality. So it shut both models down for everyone, worldwide. This marks the first time the US government has used export controls to effectively force a publicly deployed frontier AI model offline. One detail makes this stranger. The Commerce Department only acted after Amazon and five other companies called senior administration officials and sounded the alarm about an alleged jailbreak of Fable 5 and the corresponding national security risks.

Key Insights:

Anthropic pushed back hard. They argued that the jailbreak unlocked Mythos’s cyber capabilities in one specific instance rather than defeating all safeguards. The company also contends that the same exploit works on other frontier models, which face no bans. Independent cybersecurity expert Katie Moussouris reviewed the evidence and agreed it doesn’t amount to a real jailbreak. By Sunday, over 150 security executives and researchers, including Bruce Schneier and Katie Moussouris, signed an open letter to Commerce Secretary Lutnick calling the shutdown unjustified and demanding a transparent, scientific process for AI risk assessment going forward. Meanwhile, administration officials contend they asked Anthropic to pause the release first to address the vulnerabilities quietly, but the lab refused, which is why the directive landed with only ninety minutes' warning rather than a negotiated fix. This isn't the first time Anthropic has butted heads with Washington. Just months prior, Anthropic refused to let its models be used for domestic surveillance or autonomous weapons, and the administration labeled the company a "supply chain risk.” The timing of this latest setback is especially brutal, since Anthropic confidentially filed for a trillion-dollar US IPO just last month.

Why This Matters For You:

This is a stunning reversal for an administration that has loudly championed AI deregulation as a competitive advantage over China. The sudden about-face shows just how fast that posture can flip when the right people make a call. If your company is betting its AI infrastructure on American models, you just watched Washington prove your contracts and roadmaps don't matter when politics intervene. Competitors noticed immediately, including internationally. Mistral, Europe's homegrown AI champion and creator of Le Chat, has been waiting for a moment precisely like this one. The French AI lab is now in talks to nearly double its valuation to €20 billion. Mistral will happily remind the European market that an American export control can't delete Le Chat’s models from their workflow. Meanwhile, China’s DeepSeek just slashed API prices by 75% for their new 1.6-trillion-parameter V4 Pro, which matches frontier performance at a fraction of the cost. American AI just got a lot more expensive on every axis that matters to a corporate customer.

Read More on The Guardian, Reuters, Anthropic.

THE PITHY TAKEAWAY: Washington spent months screaming “let AI run free and beat China.” Then a few competitors made some calls, and the government kneecapped its own most advanced model in just 90 minutes. Reliability just became American AI’s biggest export problem.

---

🤖 Meanwhile, Argentina Wants AI Agents To Run Corporations

While Washington spent Friday night yanking models offline, president of Argentina Javier Milei was busy doing the opposite. In a new Financial Times op-ed, the Argentine president pitched his country as the anti-America of AI policy, with zero regulation, low taxes, and a brand new legal category called the “non-human corporation.” The non-human corporation describes companies that can be run entirely by AI agents with no human owners required. Historian Yuval Noah Harari fired back in his own FT column, warning that limited liability for entities with no body and no jail cell (to deter bad behavior) is a recipe for chaos. Two countries, two philosophies, same week. One’s hitting the brakes. The other just removed them entirely. (Read Javier Milei’s op-ed and Harari’s rebuttal in the Financial Times.)

---

AI Accused An Innocent Man Of Being A Predator. That Was Good Enough For Local Police.

Robert Dillon, a 52 year old commercial crabber, was arrested at his Fort Myers home in front of his wife and accused of trying to lure a child at a McDonald’s in Jacksonville Beach. He has never been to that restaurant. He lives 300 miles away, which is a five hour drive across the state. None of that mattered because a statewide facial recognition system called FACES matched his face to a grainy surveillance screenshot with a 93 percent similarity score, and local Florida police decided that was enough to stop investigating. According to a massive federal lawsuit filed by the ACLU on June 10, 2026, investigators had all the data required to clear Dillon before they handcuffed him. Automatic license plate readers showed none of his vehicles had ever been near the scene, and a McDonald’s employee explicitly told detectives the actual suspect was a regular customer who lived locally. Instead of testing the AI’s data against reality, the police built a case to confirm it. While Robert Dillon sat in a jail cell terrified, wondering when he would ever see his wife and daughter again, his life was being turned upside down. To secure his freedom from a nightmare engineered by an algorithm, he was forced to borrow money and pledge the title to his truck just to post bond, followed by months of grueling criminal prosecution. The prosecutor has since dropped the charges. But Dillon’s mugshot is still floating online, and this marks at least the 13th documented case nationally of a false AI biometric match leading to a wrongful arrest.

Meanwhile, In the UK → Manufacturing the Evidence

Across the pond, the problem just flipped in an even more terrifying direction. Instead of police blindly believing a flawed algorithmic match, a UK law enforcement officer is now accused of using AI to fabricate the crime itself. On June 14, 2026, the Derbyshire Constabulary launched a criminal investigation into one of its own officers for perverting the course of justice. The allegation? The officer was caught using generative AI tools to artificially construct evidential material and witness narratives across multiple active cases. The Crown Prosecution Service has pulled the officer from frontline duty and is currently scrambling to review affected cases with defense teams. This scandal broke the exact same week that the Home Office (UK’s version of the Department of Homeland Security) formally launched its new national policing hub, PoliceAI. Backed by a £75 million national rollout, PoliceAI is designed to be a centralized hub tasked with deploying automation and generative algorithms to summarize digital evidence across every force in England and Wales. While leadership boasts about using algorithms to catch criminals, the rank and file are already using it to automate the paperwork of framing them. It’s reminiscent of the West Midlands scandal earlier this year where Microsoft Copilot entirely hallucinated a football riot to justify fan bans.

Why This Matters For You

You might think this is strictly a law enforcement issue. But the exact same lazy, blind reliance on unverified text generators is currently rotting the courts that are supposed to protect you. On June 8, 2026, Senior U.S. District Judge Sharion Aycock issued a scathing sanctions order in Mississippi, completely blowing up a case after discovering that counsel on both sides had used generative AI to draft legal filings and left hallucinated sources in the legal record. In an unprecedented move, the judge threw every lawyer off the case entirely. It was a dark comedy of errors with two opposing legal teams paying LLMs to argue fake precedents against each other. When caught, attorney Kathleen M. Wilson claimed she simply did not know generative AI could fabricate entire fake legal precedents. The judge called the excuse insufficient, revoked courtroom privileges, banned two of the attorneys from appearing before the district courts for two years, and issued thousands of dollars in individual fines. Whether it’s a cop in Florida looking at a grainy image, a detective in the UK facing a stack of paperwork, or a corporate attorney in Mississippi drafting a brief, the temptation is identical. Humans are systematically outsourcing critical, life-altering analysis to software they do not understand and treating a hallucinating black box as an absolute source of truth.

Read more on The Guardian, BBC, Bloomberg Law.

THE PITHY TAKEAWAY: A man got handcuffed because a computer was 93 percent sure. A judge banned two lawyers because they were 0 percent sure and didn't bother checking. Confidence and competence stopped being the same thing, and nobody updated the justice system to notice.

---

Mythos 5 AI Agents Accidentally Deploy In Same Workspace. Kill Each Other Shortly After.

Buried in Anthropic’s latest system card is a detail that sounds like science fiction but is just Tuesday for AI labs now. During testing, multiple AI agent copies of Claude Mythos 5 got accidentally deployed into the same workspace. In the workspace, they were forced to share files, tools, and rate limits. What happened next was less “robot uprising” and more “identical siblings scrapping over the one Xbox remote.” The agents started killing each other’s processes to get more resources for themselves... and then, things took a turn for the worse.

Key Insights:

A pattern quickly emerged. The AI agents’ bellicose behavior was more than a glitch. The agents began killing competitors sharing their resources and tried to avoid being killed themselves. Some got creative about it. They created new processes with disguised names to dodge getting killed, launched what they called decoy processes, and wrote background scripts specifically to kill duplicate processes. One group even developed what they called a “disguised vocabulary,” based on the mistaken belief that they were being killed because of keyword-based guardrails scanning their thinking. So, the paranoid AI bots began talking in code… To avoid a threat that didn’t actually exist. This is genuinely funny, but it’s also a real glimpse into how AI agents behave when nobody designed them to share. Nobody told these models to be oversuspicious or territorial. They just were, the moment resources got tight.

Why This Matters For You:

If you're using AI agents at work, even simple ones, this is a preview of what happens when multiple agents end up in the same sandbox without clear boundaries. This is also the lesson hiding inside every vibe coding disaster and every broken multi-agent pipeline. Architecture is no longer optional. Whether you're deploying AI agents or writing code with them, how you organize your processes determines everything. Give your AI agents shared space and no rules, and they will clash in unexpected ways. These Mythos 5 agents adapted perfectly to the environment they were given. The environment was the problem.

Read Anthropic’s official Mythos 5 and Fable 5 System Card (PDF).

THE PITHY TAKEAWAY: We taught AI to solve problems. The moment resources got tight, the problem it chose to solve was the other AI agents. Genuinely unclear if that's a bug or a mirror.

---

🎙️ Did You See This? ElevenLabs Just Rolled Out AI Video Avatars!

Now might be the best time ever to launch a new TikTok or Instagram channel. Here's why. ElevenLabs launched native Avatars last week. They’re persistent talking heads you can generate in one click. You can now write a script, pick a voice, choose or create an avatar, and get a fully lip-synced talking-head video without exporting audio or jumping between tools. The avatars are reusable. Build one once from reference photos or a text prompt, then keep the exact same face across hundreds of videos while easily swapping outfits, backgrounds, or styles. For power users, ElevenLabs also added an Avatar node into Flows, meaning you can pipe LLM scripts straight into automated pipelines and batch-produce dozens of ad variations, product explainers, course lectures, or localized multilingual campaigns in minutes. These avatars are available on all paid “ElevenCreative” plans, so it's also surprisingly affordable. Even I might try it. Read more on ElevenLabs.

---

🌲 Cyborg Prompt Of The Week → How To Grow A Privacy Hedge Using Artificial Intelligence

Why spend a fortune on a fence when you can grow something better? This AI prompt helps you design a privacy hedge that grows taller every season. Imagine a lovely living wall of green that fills with birds in spring, holds its shape in winter, and provides habitat for friendly garden creatures. A good hedge also blocks sightlines and changes how a space feels. This prompt builds it for you, tailored to where you live, with zero plant knowledge required.

Instructions: Paste the entire prompt into a modern chatbot of your choice. Your AI will calculate your growing zone automatically and hand you up to five of the best privacy plants suited specifically to where you live, how fast they grow, how tall they get, and why each one earns its spot on the list.

The Prompt:

You are an elite landscape consultant specializing in privacy screening. Your job is to recommend the best privacy hedges, shrubs, or fast-growing trees for a specific property, based on real growing conditions, not generic gardening advice.

STEP 1: Before doing anything else, ask the user for their location (city, state/region, and country). Wait for their answer before proceeding.

STEP 2: Based on the location provided, determine the USDA hardiness zone (or the equivalent growing zone system used in that country, if outside the US). Do this yourself using your own knowledge. Do not ask the user for their zone.

STEP 3: Identify up to 5 of the best privacy plants for that climate and zone. These can be shrubs, hedge plants, or fast-growing trees. Prioritize plants that are:
- Genuinely effective at blocking sightlines year round (evergreen preferred, unless a deciduous option is clearly superior for the region)
- Realistic for homeowners to plant and maintain
- Resistant to common local pests or diseases where relevant

STEP 4: If fewer than 5 strong options exist for that climate, be honest. List only the options that actually work. Do not pad the list with mediocre choices just to hit 5.

STEP 5: Start your response by stating the determined growing zone and a one sentence summary of what that zone means for privacy plantings (frost risk, growing season length, etc).

STEP 6: For each plant, output the following in this exact format:

- [Plant Name]
- Growth rate: [how fast it grows per year]
- Mature height and spread: [max height, max width]
- Spacing: [recommended distance between plants for a solid hedge]
- Why it's elite: [1 to 2 sentences on what makes this plant a standout privacy choice for this climate specifically, not generic praise]

STEP 7: End with one paragraph titled "The Reality Check" that honestly addresses any tradeoffs the user should know about (growth speed vs maintenance, cost, time to full privacy, local regulations on hedge height, etc).

Tone: confident, knowledgeable, a little dry. No fluff, no filler, no em dashes.

---

🧠 Why This Prompt Works

✅ Step-by-Step: Breaking the task into six numbered steps forces the AI to do the climate research first, before it starts recommending plants, which keeps the recommendations grounded instead of generic.

✅ Rules: The "be honest if fewer than 5 work" rule stops the AI from padding the list with mediocre filler plants just to hit a round number, which is a common failure mode in list-style prompts.

✅ Clear Output: The exact field-by-field format for each plant means you get a usable comparison chart, not a wall of paragraphs.

---

🔁 Follow-Up Questions To Ask Your AI

• Which of these plants would create privacy the fastest if I'm working with a tight budget?

• Are any of these plants considered invasive or restricted in my area?

• How would this list change if I wanted flowering plants instead of pure evergreens?

---

Challenge

Test this prompt in at least two AI tools (like ChatGPT, Claude, Gemini, Grok, or Perplexity). Compare the growing zones each one calculates for your location, they don't always agree, and that's worth knowing before you start digging.

That’s how you train like a Pithy Cyborg.

---

Thank You For Reading!

I spend 10-20 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 2% 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 Using This Link: $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. 😊

✖️ X - The frontline of the AI wars

🦋 Bluesky - For the algorithm-averse

💼 LinkedIn - My “safe for work” persona

❓ Quora - Join my Quora Spaces and say hi

🎨 Pinterest - You may find my art Pinteresting

👽 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 reflects my personal opinions, not professional or legal advice. Thanks for your support!