The company racing to automate AI research is now hiring humans to figure out what happens when they succeed.
The Summary
- OpenAI posted a Preparedness team role paying $295K-$445K to prepare for AI systems that can recursively self-improve and train better versions of themselves
- The job explicitly includes tracking "progress toward automation of technical staff" — meaning the researcher would monitor how close AI gets to replacing people like them
- DeepMind's CEO just declared we're at the "foothills of the singularity", making this safety work suddenly urgent as coding agents from OpenAI and Anthropic leap forward
The Signal
OpenAI is hiring someone to stand watch at the edge of a cliff. The Preparedness safety team role isn't about theoretical risks anymore. It's about preparing for the moment when AI models start writing the next generation of AI models, without human researchers in the loop.
The job description is refreshingly blunt. You'll be reasoning about "problems that might exist in the future, but might not exist now." Translation: your job is to imagine catastrophic failure modes for technology that doesn't quite exist yet, based on systems that are evolving faster than anyone expected six months ago.
"This work relies on reasoning about problems that might exist in the future, but might not exist now."
The timing is everything. Recent leaps in coding tools from OpenAI and Anthropic have moved recursive self-improvement from sci-fi to engineering problem. When your AI can already write production code, debug itself, and reason through complex technical problems, the distance to "AI that improves AI" shrinks fast.
Google DeepMind's Demis Hassabis just said we're at the "foothills of the singularity". That's not hype. That's the head of one of the three labs closest to AGI saying the exponential curve is starting to bend. OpenAI heard that and decided they need someone tracking how close their own models get to automating technical staff.
The irony is sharp: you're hired to measure progress toward your own obsolescence. The role includes tracking automation of technical work. That's you. That's the researchers around you. The safety team needs someone tasteful and strategic enough to stay clear-eyed while building guardrails for a future where they're not needed.
The pay reflects the stakes. $295K-$445K isn't just market rate for top AI safety talent. It's compensation for cognitive dissonance. You're being paid nearly half a million dollars to develop frameworks for a world where that salary goes to zero because the AI research team is all agents.
The Implication
If you're in AI research, this job posting is a mirror. OpenAI is explicitly hiring for the transition period before recursive self-improvement. That means they believe the transition is coming, they don't know exactly when, and they want someone watching the gauges.
For everyone else, watch what OpenAI does more than what they say. Preparing for automated technical staff isn't an academic exercise. It's a signal that the agent economy isn't just automating customer service and data entry. It's coming for the people building the agents.