OpenAI: AI 80% Toward Researcher, Urges Safeguards

OpenAI says current AI systems already outperform top humans on complex intellectual tasks and are “80% of the way to an AI researcher.” The company expects AI to help make small scientific discoveries by 2026 and more significant breakthroughs by 2028, driven in part by rapidly falling “intelligence” costs (OpenAI estimates ~40x reductions per year).

Because of that trajectory, OpenAI is urging industry and governments to coordinate now on safety, oversight, and resilience for systems that could approach or exceed human-level intelligence. Recommendations include shared safety standards among leading labs, building a resilience ecosystem analogous to cybersecurity, close collaboration with regulators and safety agencies, and continuous tracking of AI’s real-world impacts to shape public policy.

The core message: timelines are uncertain, but preparation shouldn’t wait. OpenAI is attempting to set the tone for collective safeguards to reduce risks — from misuse (including biothreats) to runaway self-improvement — as models grow more powerful and their capabilities diffuse across society.

Source