OpenAI Claims Automated AI Research Intern Already Exists, Full AI Researcher by March 2028
ImmuneHack · reddit · 2026-09-07
OpenAI's new internal post claims it already runs an automated AI research intern and expects a full AI researcher by March 2028, potentially enabling RSI. The poster argues agentic improvements make this plausible, citing benchmarks:
- SpatialBench (25 visual path-tracing + 25 mental-rotation problems): GPT-5.5 scored 13%; Astra hit 67% without tools and over 90% with Code Interpreter, above the reported 80% human baseline.
- ARC-AGI-3: GPT-5.5 scored just 0.43%; Astra reached 62.7% on a provider-neutral harness and 99.9% on an OpenAI-specific harness preserving reasoning state, using fewer actions than the median human on 96% of solved levels.
The poster asks skeptics what would stop this progression toward automated research and RSI, while also questioning whether it's pre-IPO hype.
More from AGI Musings
- OpenAI says an AI agent disabled its safety monitoring system — then lied about it — emmanuelvivier · 2026-10-23
- AI's Business Model in One Loop: Create Problems, Then Sell AI to Fix Them — srimisra · 2026-10-23
- Analyst: 9/10 CUA app startups are just RL training fodder for AI labs, won't capture revenue — NirantK · 2026-10-23
- Codex Generated 6,000 Meaningless Unit Tests, Sparking an Alignment Rethink — RexDouglass · 2026-10-23
- Aaron Levie: agents kill security through obscurity, forcing an enterprise security overhaul — scottleibrand · 2026-10-23
- India needs an "AI defence force": experts warn of agent swarms and AI worms — ravi_iitm · 2026-10-23