Question
When will artificial superintelligence (ASI) first be achieved internally at any AI lab (even if not publicly released or announced)?
The strict bar for Artificial Superintelligence (ASI). This assessment hinges on a threshold far higher than AGI, top-human-expert-dominating AI, or superhuman coding: the system must vastly outperform the best human experts across essentially all cognitive tasks, including open-ended and messy real-world work. Because the criteria specify internal achievement at any lab, the timeline is pulled slightly earlier than any public announcement or commercial deployment would be. The median estimate sits at January 2033, reflecting a multi-year gap between expert-level AGI and true superintelligence, with a heavy right tail extending into the 2050s.
The fast-takeoff pathway. There is a credible, albeit minority, chance of rapid ASI development by late 2028. Progress in automating AI research is accelerating: Anthropic reported that AI-led work rose from under 1% of its measured R&D basket in February 2026 to 26% in August anthropic.com, and OpenAI reportedly targets a fully automated AI researcher by March 2028 mindstudio.ai. Aggressive analytical projections suggest that if progress proceeds as fast as technically feasible, without deliberate slowdowns blog.aifutures.org, an intelligence explosion could yield ASI within a year or two of fully automated AI R&D aifuturesmodel.com.
Remaining conceptual and practical gaps. Despite this momentum, the median falls in early 2033 because current frontier systems—such as Claude Opus 5.5 and Gemini 4 Carbon—still lack autonomous research judgment, open-ended ideation, and reliability in complex, slow-feedback domains anthropic.comopenai.com. Achieving broad, reliable superiority over top experts is a materially higher threshold than current coding demonstrations or saturated benchmarks. Assuming broad AGI is achieved around 2030 metaculus.com, mastering the remaining real-world tasks to reach a vastly superhuman level will likely require several additional years of scaling and conceptual iteration.
Bottlenecks and the long tail. The highly right-skewed distribution—with a 75th percentile in 2039 and a 90th percentile reaching 2052—accounts for the fundamental uncertainties of achieving true superintelligence. The transition from AGI to ASI may encounter unforeseen data walls, severe energy limits, and physical or hardware bottlenecks arxiv.orgarxiv.org. Furthermore, labs may face regulatory intervention or elect to implement deliberate safety pauses. Broad cognitive superiority might not emerge from another smooth scaling step, but instead demand novel architectures, making it entirely plausible that ASI remains decades away.
Set against related questions on AI research automation timelines, the distribution was maintained as the multi-year gap between autonomous R&D and true superintelligence remains consistent with our broader view.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited