Scott Aaronson Recants Two Decades of AI Skepticism, Says the Singularity Has Begun

Scott Aaronson, the quantum computing theorist and former OpenAI alignment researcher, published a blog essay on September 16, "The Age of Wonders and Terrors," publicly retracting his twenty years of AI-risk skepticism. He conceded his past doubts about "AI takeover" scenarios were wrong, stated that Eliezer Yudkowsky was right that AI is civilization's greatest challenge in our lifetimes, and declared that the technological singularity has already begun. The essay spread rapidly across X, with Ethan Mollick, Gary Marcus and others joining the discussion; Marcus pushed back against the singularity claim.

Confirmed

• Aaronson recalled that twenty years ago, mainstream CS scholars including himself found scenarios of a recursively self-improving superintelligence erupting from a basement implausible, and even conservative peers shared that skepticism, which was then seen as reasonable caution.
• According to multiple sharers including @paraschopra, @nanjiangcs, @tokenbender and @beenkim, he admitted that the AI warning signs he and peers once dismissed have now come true one by one.
• Per @FlorianGallwitz, Aaronson stated bluntly that on the question of our lifetime's greatest challenge to civilization, Eliezer Yudkowsky was right and both he and the reader were wrong.
• Per @austinc3301, the essay invokes rats, Yudkowsky and a road-to-Damascus conversion, and clearly asserts the singularity has started; @birchlse's post corroborates this.
• Per @avtim, the essay also discusses AI's present reality as an age where wonders and terrors coexist; @connoraxiotes relays a personal note of visceral unease felt while putting children to bed, and a call to stop grinding ideological axes and trading blame.
• Per @GaryMarcus, Nate Witkin criticized the essay as detached from reality, citing a year of knowledge-work benchmark curves to dispute claims of steep month-over-month capability gains.

Unconfirmed

• Ethan Mollick's follow-on policy question: if norms of open sharing at AI labs keep eroding, will labs hoard knowledge to avoid PR crises. This is Mollick's concern, not a verified conclusion from Aaronson's essay.

Why it matters

• Aaronson is a prominent scholar who long held a skeptical position; his public reversal is a landmark moment for the AI-risk discourse and lends mainstream backing to Yudkowsky's long-marginalized warnings.
• The rebuttals from Marcus and Witkin, backed by benchmark data, show the debate over whether capabilities are surging and whether the singularity has begun is far from settled.
• Mollick's question about knowledge hoarding points to a structural tension between transparency and secrecy in AI governance, a policy issue worth tracking beyond the essay itself.

2026-09-16 ~ 2026-09-17 · 16 related posts

• Scott Aaronson: the AI warning signs we demanded have all come true — nanjiang_cs · 2026-09-16
• Scott Aaronson publishes a new post and the AI community takes notice — john__allard · 2026-09-16
• [source] Scott Aaronson: the age of AI wonders and terrors is here — will labs start hoarding knowledge? — emollick · 2026-09-16
• Scott Aaronson says the Singularity has already started in new post on Yudkowsky — austinc3301 · 2026-09-16
• Scott Aaronson says the singularity has begun — birchlse · 2026-09-16
• [source] Gary Marcus rebuts Scott Aaronson's singularity claim: benchmark curves show no monthly capability explosion — GaryMarcus · 2026-09-17
• "The wonders and terrors are here": writer on feeling dread putting his kids to sleep — connoraxiotes · 2026-09-17

9 near-duplicate retellings: avt_im · paraschopra · FlorianGallwitz · soumitrashukla9 · tokenbender · _beenkim · PeterBowdenLive · MoonL88537 · connoraxiotes