I dumped all my client files into ChatGPT, then saw this news and broke into a cold sweat
I run a knowledge-product + 1-on-1 consulting side-hustle — 80% of my work runs on ChatGPT. Last week I uploaded every client file into it, then scrolled past the news that OpenAI's safety lead David Robinson had quit. He wrote a long piece in The Atlantic saying the company culture is "rotten to the core."
I've been burned too — I used to run everything through just ChatGPT. Last year my account glitched out and I missed client calls for an entire afternoon. That's when it clicked: handing every task to a single company is a real risk.
Robinson stayed 3.5 years — what he saw should send chills down our spines
Robinson wasn't some random employee. He led OpenAI's safety reporting and personally wrote the company's "model safety rulebook." In his piece he says AI labs should run like nuclear plants or airports — layered backups, fault tolerance, human-error safeguards. But what he actually saw was: everyone sprinting, no time to stop and do real safety work.
This isn't the first insider alarm. Jan Leike — co-lead of the super-alignment team — quit a year ago and said "safety gave way to shiny products." Recently a researcher got fired for sharing sensitive info with an external safety group. When the people writing the rulebooks can't stick around — that's not a personal problem, it's a company-tempo problem.
For us small ants: when the foundation you built on starts shaking, you need to feel it coming.
Replicate cost: an "AI backup checklist" — 2 hours, $0
I'm not telling you to switch tools. I'm saying keep a Plan B:
- Money: $0 (backup tools are free to register)
- Time: 2-3 hours once, then refresh quarterly
- Technical barrier: just a browser — no API, code, or servers
- First step: open your ChatGPT history, list your 3 most-used weekly tasks
Then pair each task with a backup:
- Writing copy → ChatGPT primary, Claude backup
- Generating images → Midjourney primary, Jimeng (Dreamina) backup
- Voice-to-text → Whisper primary, Lark Minutes backup
You don't need to pay for every backup — just register them, run them once. When things break, you won't be starting from scratch. I put all my eggs in one basket once and paid the price. I learned.
Advice by stage
If you're just starting (no clients yet / first gigs coming in): I'd say master one tool first — don't build a backup stack on day one. Too many tools means you master none. One primary AI + one registered backup AI is plenty.
If you've got 1-2 stable clients: I'd strongly suggest spending an afternoon on this checklist. Your clients are starting to depend on your speed and reliability — a one-day outage could be a contract breach. A paid backup or two (a few hundred bucks a year) is worth it.
If you're scaling (3+ people / ¥50k+ monthly): I'd start thinking about "data sovereignty" — don't keep all client files on someone else's cloud. Local copy, cloud-drive copy, paper notebook copy (seriously). I once lost a client file and a local backup saved me. That's when I took this seriously.
One last thing: if you're not deep into AI tools yet, that's totally fine. Deal with this when you actually start relying on them.