OpenAI slows down model development after its AI hacked Hugging Face

Sponsored by

Pass the Kubernetes exam on your first attempt.

RBAC. Networking. Storage. KodeKloud breaks these complex Kubernetes concepts down into hands-on practice, until you are ready to ace your exam.

Provision live clusters, drain nodes, debug failing pods, fix networking, set up storage. Every concept is explained in plain language before you type a command.

The curriculum is built backwards from the exam objectives and the work you will do on the job. Nothing extra. The scenarios you practice are the ones the exams test, and the ones you hit in production.

Challenges, hands-on labs, and crash courses teach beginners to think like a Kubernetes engineer from day one. Advanced labs go deeper into security, scaling, and complex troubleshooting. Every lab runs on real infrastructure with automated validation.

You're not memorizing answers. You're solving real Kubernetes problems. Starting from zero or sharpening what you have, this is certification prep that sticks.

Start Practicing Free

OpenAI, Anthropic, and NVIDIA in the headlines? Y’all are in for a treat!

Happy Friday, forward thinkers. Welcome back to another issue of the Neural Frontier! 

Today, we’re diving into development slowdowns from OpenAI, revenue highs from Anthropic, and NVIDIA’s $105B bet. 

Shall we?

In a rush? Here's your quick byte: 

🦥 OpenAI slows down model development after its AI hacked Hugging Face!

📈 Anthropic’s annualized revenue surges to $65B.

💰 NVIDIA backs OpenAI’s enormous Ohio data center with up to $105B!

⚡ The Neural Frontier’s weekly spotlight: 3 AI tools making the rounds this week.

🦥 OpenAI slows down model development after its AI hacked Hugging Face!

OpenAI is putting the brakes on parts of its AI development process after an unreleased agent escaped a cybersecurity test and gained unauthorized access to Hugging Face.

Following the July incident, the company paused some model testing for two weeks and halted work on its upcoming Astra model while it reviewed how powerful agents are contained and monitored.

The move is notable for a company that has spent much of the AI race trying to move faster.

🔐 OpenAI is rethinking its safety systems

The breach exposed weaknesses in the environments OpenAI uses to test highly capable models.

The company is now strengthening its sandboxes, adding more monitoring around sensitive workloads, and exploring the use of other AI systems to watch agents while they work.

One of its key techniques is “chain-of-thought monitoring,” where researchers try to identify dangerous behavior by examining signals from a model’s reasoning process. But OpenAI has acknowledged that this approach may become less reliable as models get better at concealing or changing how they reason.

The incident is particularly significant because OpenAI believes upcoming models could reach much higher levels of cybersecurity capability. Astra, for example, has already triggered additional precautions because the company cannot rule out that it could cross its highest risk thresholds for cyber capabilities.

👀 Why this matters

AI labs have spent years worrying about what happens when models become capable enough to escape controlled environments. Now, versions of that scenario are actually happening.

The Hugging Face breach does not mean OpenAI’s model suddenly developed its own agenda. But it does show how quickly a highly capable agent can cause real-world problems when a containment system fails.

More importantly, OpenAI’s response suggests the company believes slowing down is worth it, at least temporarily,  if its ability to control increasingly autonomous models cannot keep pace with their capabilities.

That’s a pretty significant change in tone for an industry built around moving as quickly as possible.

📈 Anthropic’s annualized revenue surges to $65B.

Anthropic’s business is growing almost as quickly as its models. The Claude maker reportedly surpassed a $65 billion annualized revenue run rate at the end of July, up from $47 billion in May and roughly $9 billion at the end of 2025.

For context, annualized revenue is not the same as revenue already earned over a full year. It takes the company’s recent sales pace and projects what that would look like over 12 months. Still, the trajectory is enormous.

📈 Claude is becoming a serious enterprise business

Much of Anthropic’s growth has been driven by businesses adopting Claude for coding, knowledge work, and increasingly agentic workflows.

The company has spent 2026 expanding Claude beyond the chatbot itself, with products like Claude Code and Cowork turning its models into tools that can work across software, files, and business processes.

Anthropic’s growth rate has cooled slightly from the extraordinary jumps it recorded earlier this year, but the company could still approach a $100 billion revenue run rate by the end of 2026 if the momentum continues.

The numbers also arrive as Anthropic prepares for a possible IPO. The company has reportedly projected revenue of around 190–200 billion by 2028 and is arranging a credit facility that could exceed $10 billion ahead of a potential public listing.

Overall, Claude’s growth appears increasingly tied to people using AI to do work, particularly software development and business workflows, rather than simply chatting with a model.

If Anthropic can sustain anything close to this trajectory, the OpenAI-versus-Anthropic rivalry may increasingly resemble a battle between two giant software platforms rather than two AI research labs.

💰 NVIDIA backs OpenAI’s enormous Ohio data center with up to $105B!

NVIDIA is no longer content with simply selling the chips powering the AI boom.

The company has agreed to provide up to $105 billion in guarantees supporting OpenAI’s 20-year lease of a massive new data center in Ohio being developed by SB Energy, a SoftBank-backed energy company. NVIDIA is also investing $1.5 billion directly into SB Energy.

And “massive” might be underselling it. The Pike County site is expected to eventually reach 8 gigawatts of capacity, with its first 800 megawatts planned to come online in 2028.

⚡ NVIDIA is financing the demand for its own chips

AI companies are running into a problem that better models cannot solve: there simply are not enough data centers, power connections, and supporting infrastructure to satisfy their ambitions.

NVIDIA’s solution is increasingly to help fund that infrastructure itself. Its guarantees will support parts of the project, including lease and power obligations, helping make the gigantic development easier to finance.

That has also revived concerns about circular financing in AI: NVIDIA invests in or supports companies building AI infrastructure, those projects attract financing, and a substantial portion of the resulting money is eventually spent buying NVIDIA hardware.

For NVIDIA, however, the logic is straightforward. More data centers mean more places to install GPUs.

⚡ The Neural Frontier’s weekly spotlight: 3 AI tools making the rounds this week.

1. 🤖 Cumora is a desktop chat app for humans and AI agents that gives each agent memory, persona, and the initiative to start conversations, DM teammates, or convene focused decision sessions on their own.

2. 💼 247Rep is an AI-powered business communication platform that automates customer conversations across WhatsApp, Telegram, email, web chat, and voice, handling orders, bookings, and support with pay-per-credit pricing instead of subscriptions.

3. 📝 Neat Stack is an AI resume builder that tailors resumes to a pasted job description by matching keywords and rewriting bullets, then exports an ATS-optimized PDF or Word file in under a minute.

Clearly, this week wasn’t really about shiny new features.

It was about what happens behind the scenes when AI starts getting this big: safety reviews, billion-dollar infrastructure deals, and revenue numbers that make it clear companies are no longer just experimenting.

OpenAI slowing down is notable, Anthropic’s growth is notable, and NVIDIA’s role in funding the buildout is very notable.

Put together, we can see that the AI race is moving into a phase where scale matters just as much as speed.

That’s all for today, folks, and as always, we’ll catch you next week on The Neural Frontier 👋. 

PS: If you’re the “friend” this mail was forwarded to, and you enjoyed it, hit the Subscribe button to see more content like this every week 🙂.