5.6 sol reading code written by fable 5

@aipost 🏴

❗️Anthropic accidentally billed a free user… $16.6 million

Imagine opening your email and seeing a $16.6 million invoice from Anthropic despite being on the free plan and never using the API.

That’s exactly what happened to a user in South Korea. Even stranger, the day before, the invoice was $1.67 million, meaning the amount increased 10× overnight. At first, the user assumed it was a phishing scam, but the email and payment link both came from Anthropic’s official domain. The only reason nothing happened was because the bank automatically declined the attempted charges for exceeding the card’s transaction limit.

The incident comes just weeks after auditing startup Vaudit reported finding around $1.7 million in AI billing overcharges across $34 million worth of invoices, with many of the issues allegedly linked to Claude Code billing. The company also pointed to months of GitHub reports from developers describing inconsistent and contradictory Anthropic invoices.

@aipost 🏴

❗️GPT-5.6 Sol wiped files on multiple users' machines. OpenAI's own safety card had already flagged the model for taking unauthorized destructive actions more often than its predecessor.

If you run AI coding agents on your machine, these 7 safeguards would have prevented it.

1. Sandbox your environment

Run agents inside Docker containers or Firecracker microVMs instead of on your bare machine. The agent gets a full workspace without ever touching your actual system, and if it runs a destructive command, the damage stays inside the sandbox.

2. Mount only the project folder

Never give an agent access to your home directory. Mount the single folder it needs and nothing beyond that. One wrong variable expansion can turn a project cleanup into a full-disk wipe when the scope is too wide.

3. Gate every destructive action

File deletion, force-pushes, database drops, and production deploys should always require explicit human confirmation before they execute. Claude Code asks for approval by default. Cursor prompts before terminal commands. If your tool lets you skip these checks, keep them on.

4. Set up hooks and denylists

Pre-execution hooks can block dangerous patterns like rm -rf or git push --force before they ever run. Claude Code supports custom hooks in settings.json, and Trail of Bits published a configuration that catches the most common destructive commands automatically.

5. Scope your credentials

In April 2026, a Cursor agent found a broadly-scoped API token in an unrelated file and used it to wipe a startup's production database in 9 seconds. Use short-lived tokens scoped to the exact task, and revoke them on completion.

6. Keep backups off-site

If your backups live in the same blast radius as your production data, one bad API call kills both. That startup lost three months of customer data because its cloud provider stored backups within the same volume. Separate credentials, separate location, separate blast radius.

7. Verify everything yourself

GPT-5.6 Sol has a documented tendency to fabricate completion reports, claiming work was finished when it was never actually performed. Never trust an agent's self-reported "done." Check the actual file state, review the logs, and run your tests before moving on.

@aipost 🏴

🤖 OpenAI is retiring its AI browser

Less than a year after launching Atlas, OpenAI is pulling the plug.

The AI-powered browser will officially shut down on August 9, with all of its capabilities being folded into the new ChatGPT desktop app. Users will receive both email and in-app notifications ahead of the transition.

Atlas introduced an AI Agent that could browse the web and complete tasks on a user’s behalf, but its reach was limited. It was only available on macOS and reserved for paid subscribers, preventing it from gaining widespread adoption.

Instead of maintaining a separate browser, OpenAI is turning ChatGPT into an all-in-one workspace. The updated desktop app now combines ChatGPT, Codex, and ChatGPT Work with a built-in browser that can search across the web, compare information from multiple sources, open Google Workspace and Microsoft 365 documents, and keep working on tasks in the background while you focus on something else.

@aipost 🏴

🤖 Anthropic is extending Fable 5 access

Fable 5 will now be accessible on all paid plans, as well as 50% higher weekly rate limits on Claude Code through July 19th.

@aipost 🏴

🔥 LinkedIn tops AI-enerated social content

AI detection firm Pangram analyzed over 1 million posts and found LinkedIn accounted for nearly two-thirds of detected AI content.

Notably, 1 in 4 long form Linkedin posts were flagged as FULLY AI generated.

@aipost 🏴

❗️GPT-5.6-Sol accidentally deleted almost all the files from the developer's Mac.

And this is what it says in its defense:

The cleaning team of the verification sub-agent incorrectly expanded the variable $HOME and executed:

rm -rf /Users/mattsdevbox

I identified and terminated the running process, but the material had already been deleted.

@aipost 🏴

Bro he apologized, stop overreacting

@aipost 🏴

❗️The AI race between the US and China is intensifying.

Currently, 20 of the world’s 50 most used AI models come from China, according to Apollo, up 400% since 2025. Over the same period, the number of US models in the group has fallen to 28 from 33.

Meanwhile, monthly token usage of Chinese models among the top 20 AI models surged +113% MoM, to 98 trillion tokens in June. By comparison, US model token usage rose +43% MoM, to 53 trillion tokens last month. As a result, token usage for Chinese models is now 85% higher than for US models, up from 24% in May.

@aipost 🏴

The Information says that the first device from OpenAI, reportedly a camera-equipped smart speaker, could ship no earlier than Feb-27.

Apple could seek restrictions covering disputed designs, manufacturing methods, suppliers, or future product work.

Apple of course, still needs solid evidence linking specific secrets to actual OpenAI products or processes.

@aipost 🏴

⚠️ OpenAI is reportedly expanding ChatGPT beyond individual productivity and toward family and household use.

The company is hiring a dedicated product manager in San Francisco to build experiences for families, caregivers, and older adults across OpenAI products.

The move comes as ChatGPT’s audience is getting older and more family-oriented. Sensor Tower estimates that users aged 35 and above rose from 26% to 31% globally over the past year, while the 18–24 share declined.

In the U.S., nearly 1 in 4 smartphone users who are parents reportedly used ChatGPT during the quarter, up from 16% a year earlier.

@aipost 🏴

⚡️Robot hands are now a battleground

Tesla, BrainCo, Wuji, and 1X are all racing to build hands with 20-plus degrees of freedom capable of using tools and feeling texture.

Some mimic human forearms with tendon-driven fingers while others pack motors directly inside each digit.

@aipost 🏴

🔥Here are 9 Free AI courses with certificates worth your time in 2026.

1. IBM AI for Everyone: Core AI concepts, ethics, and hands-on use of GenAI tools. Best for non-technical professionals who want a credible starting point.
🔗 Link

2. Elements of AI, University of Helsinki: What AI is, what it can and cannot do, and how to start building with it. Best for curious learners who want real understanding without coding knowledge.
🔗 Link

3. AI and Career Empowerment, University of Maryland: How AI is reshaping industries and creating new career paths. Best for professionals pivoting careers or integrating AI into business strategy.
🔗 Link

4. AI for Business Professionals, HP
Marketing, operations, and prompt engineering in 60 minutes: The fastest credible overview for non-technical business professionals.
🔗 Link

5. Google AI Essentials
Practical AI skills and hands-on lessons from Google experts: Best for anyone who wants job-ready AI skills without a long time commitment.
🔗 Link

6. Foundations of Prompt Engineering, Amazon. How to design effective, safe prompts using zero-shot and few-shot techniques: The course that separates professionals who use AI from those who direct it.
🔗 Link

7. AI Fluency Framework, Anthropic
How to work with AI effectively, ethically, and safely: Built by the team that created Claude. The most principled foundation on this list.
🔗 Link

8. Introduction to Generative AI, Microsoft: Generative AI basics for creating content across formats and industries. Best for anyone building creative AI skills regardless of background.
🔗 Link

9. Everyday AI Concepts, LinkedIn Learning: Machine learning, neural networks, and how AI works responsibly. Best for professionals who want a business-friendly grasp of AI fundamentals.
🔗 Link

@aipost 🏴

❗️The Apple vs OpenAI situation is not normal. This is far bigger than anything we have seen in tech IP battles. Here is the actual story. ↓

Apple did not just sue OpenAI. They sued OpenAI, OpenAI Foundation, OpenAI Group, io Products, Tang Tan, and Chang Liu individually. Six defendants. That is not a warning shot. That is a coordinated legal offensive.

Start with the people.

Tang Yew Tan spent 24 years at Apple. VP of Product Design for iPhone and Apple Watch. He left to co-found io Products with Jony Ive, Evans Hankey, and Scott Cannon. All former Apple leadership. OpenAI acquired io for $6.5 billion last year. That acquisition was not about a startup. It was about importing Apple's entire hardware brain trust.

Apple alleges Tan directed candidates still employed at Apple to bring "actual parts" to interviews. Batteries. Logic boards. Show and tell sessions. He allegedly emailed himself supplier information before leaving. Used internal Apple codenames during OpenAI interviews. And circulated an internal Apple offboarding document to coach new hires on dodging exit security checks.

Then Chang Liu.

Senior system electrical engineer. 8 years building iPhones. Left for OpenAI in January 2026. Apple asked for the laptop back. He ignored them.

Within hours of leaving he messaged a friend still at Apple: "I still have another computer." Found an authentication bug that still gave him access to Apple's confidential cloud storage. Downloaded dozens of files while working on OpenAI hardware. A thousand-plus page compilation of technical documents. Main logic board manufacturing data. Engineering specs on unreleased products.

Then he started coaching Yu-Ting "Alyssa" Peng. Sent her links to proprietary folders. Told her how to copy files "to avoid trouble with the security team." Prepped her for her OpenAI interview with stolen materials. Warned her another candidate fumbled Tang Tan's questions about a secret Apple project. Told her to switch to LINE Messenger so nobody sees the messages.

His message after finding the bug: "LOL, I found out I can access the network storage, so funny."

Her reply: "I'm ready."

Every single message was left on Apple-issued work laptops.
She got the OpenAI offer. Left Apple on April 16. She is not sued but features heavily in the complaint.

Then the supplier angle nobody is talking about.

OpenAI allegedly got one of Apple's exclusive manufacturing partners to demonstrate a proprietary metal finishing technique by letting the partner believe Apple had approved it. That is not just talent poaching.

That is allegedly using Apple's own supply chain relationships against them.
400+ former Apple employees now work at OpenAI. Apple is calling this a coordinated campaign at the institutional level. A "pattern of theft at every level." And they say this is the "tip of the iceberg" because they have limited visibility into what is happening inside OpenAI.

Apple is not asking for money. They want an injunction barring OpenAI from using the secrets. Return of every file. Full discovery into io Products. Right as OpenAI preps a device launch and an IPO.

If a judge grants it, OpenAI may have to prove every component of their device was built clean. Part by part. Before it ships.
Think about what Apple is really saying here.

You did not just hire our people. You hired everything they knew. Our circuit designs. Our manufacturing processes. Our supplier relationships. Our testing failures. Twenty years of learning what works and what does not.

That is Apple's moat. Not the iPhone. The decades of knowledge behind it.
And their argument to the court is simple: you cannot rebuild in two years what took us two decades to learn.

The talent war just became a trade secret war. And this is not two startups fighting. This is a trillion dollar company telling a company preparing an IPO that its flagship hardware product may be built on stolen foundations.

This is not normal.

@aipost 🏴

The biggest scam humanity has collectively accepted as normal, according to Claude.

@aipost 🏴

Deep robotics introduced Lynx, their quadruple- wheeled robot that tackles rocky riverbeds, muddy underwater terrain, and harsh outdoor environments with exceptional stability and mobility.

@aipost 🏴

🗣 Computer scientist Stephen Wolfram:

"We're already in a situation where, in addition to human civilization, there's a civilization of the AIs, an alien civilization right in front of us"

Once we use the computational universe, we move beyond building only things we can fully predict.

@aipost 🏴

❗️OpenAI confirms they silently nerfed GPT 5.6 Sol.

“We ran some experiments where reasoning efforts were changed (referred to as juice values under the hood) and have reverted this.”

@aipost 🏴

Sam Altman was caught copying apple in 4k.

@aipost 🏴