Safety For Hire: OpenAI Offers $200K to Outsiders After Quietly Dismantling Its Entire Internal Safety Division

In what might be the most aggressively ironic move of the quarter, OpenAI announced a Safety Fellowship on April 6 — a program offering external researchers $3,850 per week, plus roughly $15,000 in monthly compute and mentorship — just hours after The New Yorker published an investigation detailing how the company had dissolved three consecutive internal safety teams over 22 months. It's giving "closing the stable door after the horse has already emigrated to a non-extradition jurisdiction" energy, if we're being honest.

The fellowship, described as "a pilot program to support independent safety and alignment research," is not small change. At over $200,000 annualized, plus compute resources and mentorship from OpenAI researchers, it's a genuine investment in external brainpower. Applications close May 3. Fellows work from Constellation's Berkeley workspace or remotely, and they're not limited to AI specialists — OpenAI is recruiting across cybersecurity, social science, and human-computer interaction alongside computer science. Basically, they're crowdsourcing the safety department like a DAO governance proposal, except nobody gets voting rights.

But here's the rub: fellows get API credits and compute resources. No system access. It's arm's-length research funding, not a rebuild of the internal infrastructure that just got dismantled. Imagine hiring external auditors to review your DeFi protocol's security while permanently revoking their access to the codebase. Technically you're funding research. Practically you're just buying a really expensive PR buffer.

The Timeline Tells the Story

Ronan Farrow's investigation in The New Yorker documented the dissolution of three internal safety organizations in swift succession. The superalignment team — co-led by Ilya Sutskever and Jan Leike — was shut down in May 2024. Leike's exit message was notably pointed: "safety culture and processes have taken a backseat to shiny products." The AGI Readiness team followed in October 2024. The Mission Alignment team, which lasted just 16 months, was disbanded in February 2026. For context, that's three rug pulls in under two years. The safety teams, not the products.

The New Yorker also reported that when a journalist asked to speak with OpenAI's existential safety researchers, a company representative replied: "What do you mean by existential safety? That's not, like, a thing." Peak vibes. Very "I don't recall" energy, but for the apocalypse.

What Fellows Must Produce

The research agenda spans seven priority areas: safety evaluation, ethics, robustness, scalable mitigations, privacy-preserving safety methods, agentic oversight, and high-severity misuse domains. Each fellow must produce a substantive output by February 2027 — a paper, benchmark, or dataset. Specific academic credentials aren't required; OpenAI says it prioritizes research ability, technical judgment, and execution capacity.