What happened to OpenAI's alignment team?

In May 2024, roughly 50% of OpenAI's Superalignment team departed, including team leader Jan Leike and co-founder Ilya Sutskever. Leike cited concerns that 'safety culture and processes have taken a back seat to shiny products.' The team was originally created to ensure advanced AI systems remain aligned with human values.

Why does AI safety matter for everyday users?

AI safety affects data privacy, reliability, and trust. When safety-focused researchers leave a company, it raises questions about how that company handles user data, how thoroughly they test new features, and whether they're prioritizing user interests over rapid product releases.

What is local AI and how is it different from ChatGPT?

Local AI runs directly on your device (like your Mac or iPhone) rather than on remote servers. This means your documents and prompts stay on your computer, offering stronger privacy protection. ChatGPT and similar tools process everything on cloud servers, where your data is subject to the company's policies.

How does Elephas protect my privacy?

Elephas offers local AI processing through Ollama integration, meaning your documents never leave your device. Even when using cloud models, Elephas gives you control over what data gets shared. Your Super Brain knowledge base stays private, and your data is never used to train AI models.

AI PrivacyFebruary 15, 2026 · 8 min read

OpenAI's Safety Team Exodus: Why Local AI Matters More Than Ever

Something troubling happened at OpenAI that every AI user should know about. In May 2024, Jan Leike resigned from OpenAI's Superalignment team—the group specifically tasked with making sure AI stays aligned with human values. His parting words were blunt: “Safety culture and processes have taken a back seat to shiny products.”

50%

Superalignment team departed

20%

Computing promised but never delivered

Co-founders departed over safety

100%

Your data stays local with Elephas

Executive Summary

Jan Leike resigned from OpenAI's Superalignment team citing safety taking “a back seat to shiny products”
Ilya Sutskever, OpenAI co-founder and Chief Scientist, also departed the company
Roughly 50% of the Superalignment team left—the group tasked with keeping advanced AI aligned with human values
The team was promised 20% of OpenAI's computing resources but priorities shifted to product releases
Similar departures at Anthropic, with researchers warning “the world is in peril”
For knowledge workers, this raises critical questions about data privacy and trust in cloud AI

What Actually Happened at OpenAI

OpenAI created the Superalignment team to solve one of AI's hardest problems: how do we make sure advanced AI systems remain helpful and don't cause harm? It was supposed to receive 20% of the company's computing resources. The team attracted some of the brightest researchers in the field.

Then things changed.

As OpenAI raced to stay ahead in the AI arms race—competing with Anthropic, Google, and a flood of startups—the balance shifted. According to departing team members, the pressure to release new products started winning out over the slower, more careful work of safety research.

Jan Leike's resignation letter put it plainly. He wrote about the constant tension between shipping features and ensuring those features won't cause problems down the line. When push came to shove, shipping won.

This pattern isn't unique to OpenAI. We've seen similar departures at Anthropic, with safety researchers like Mrinank Sharma warning that “the world is in peril.” Engineers at multiple AI companies have expressed concerns about the gap between what AI can do and what safety teams can control.

Why This Matters for Knowledge Workers

You might be thinking: “I just use ChatGPT to help draft emails and summarize documents. What does alignment research have to do with me?” More than you'd expect.

The Data Question

When you use cloud-based AI tools, your prompts, your documents, and your intellectual property travel through someone else's servers. OpenAI's terms of service have changed multiple times. Data retention policies vary. And if a company's safety team—the people who care most about responsible AI use—is walking out the door, what does that tell you about how seriously they're taking other concerns?

The Trust Gap

AI tools see some of our most sensitive work: legal documents, financial plans, creative projects, health concerns. We're placing enormous trust in companies that are increasingly prioritizing “shiny products” over careful development. The researchers leaving these companies aren't junior employees—they're the ones who understood the risks best.

The Dependency Problem

Many knowledge workers have built AI deeply into their workflows. What happens when the company behind your favorite AI tool pivots, changes pricing dramatically, or gets acquired? The centralized, cloud-dependent model creates fragility.

When the safety experts head for the exits, everyone using these products should ask why.

The Case for Local-First AI

This is where the model matters. Local AI keeps your data on your device. It doesn't require your documents to travel to servers you don't control. It works offline. And it puts you in charge of the trade-offs between convenience and privacy.

The trend toward local AI isn't just about paranoia—it's about control:

Why Local AI Wins

Privacy by architecture: When processing happens on your device, there's no server to breach, no data retention policy to worry about, no terms of service that might change tomorrow.
Independence from corporate decisions: If a cloud AI provider decides to change their model, their pricing, or their approach to safety, you're along for the ride. Local AI gives you options.
Reliability: Local AI works on a plane, in a coffee shop with spotty wifi, or when the cloud service is having an outage.

How Elephas Approaches AI Differently

At Elephas, we've built a personal AI assistant specifically for knowledge workers who care about where their data goes. Here's what that means in practice:

Local Processing with Ollama Support

You can run AI models entirely on your Mac. Your confidential documents, your client information, your private thoughts—they never leave your device if you don't want them to.

Multi-Provider Flexibility

Choose between cloud models (OpenAI, Claude, Gemini) when you need maximum capability, or switch to local models when privacy matters most. You're not locked into any single provider's decisions.

Super Brain: Your Documents, Your Answers

When you need AI that understands your specific work, Super Brain creates a personal knowledge base from your own files. Every answer includes citations—no hallucinations, no made-up facts. Just your information, organized and accessible.

System-Wide Access

One keyboard shortcut brings AI to any Mac app. No switching tabs, no copying and pasting.

The point isn't that cloud AI is bad. It's that you should have choices. And when the safety teams at major AI companies are raising red flags, having a privacy-first option isn't paranoia—it's prudent.

Try Elephas Free — Starting at $9.99/month

What This Means Going Forward

The AI industry is at an inflection point. The technology is remarkable, but the institutions building it are under pressure to move fast and take risks. The people who understand those risks best—the safety researchers—are increasingly choosing to leave.

For knowledge workers, this creates a choice: continue trusting that cloud AI companies will prioritize your interests, or take more control over your AI tools.

Neither choice is wrong. But the choice should be informed. And right now, the people inside these companies are telling us to pay attention.