Status as of October 2026. Despite an extraordinary surge in AI mathematical capability, none of the nine target problems has a full, accepted resolution. Recent major breakthroughs all fall short of the required full statements: the Navier–Stokes proof covers only the forced case (Clay statements C and D) On the Navier–Stokes Millennium Prize Problem [Existence and Smoothness of the Navier–Stokes Equation](https://www.claymath.org/wp-conte
To ensure strict coherence with a related probability estimate , the 90th percentile of this timeline was pushed out to 2040-01-01 to reflect a 10% chance of no problem being resolved before 2040.
AI systems are now rapidly producing major mathematical results, including a Lean-checked forced Navier–Stokes blowup AI Has Solved One of Math's $1 Million Millennium Prize ..., an ETH refutation, and Yang–Mills/QCD existence proofs OpenAI Says Its Models Produced Three Proofs in Math .... However,
Set against the timelines for pure mathematics breakthroughs , this distribution was shifted slightly later to account for the historical lag required for theorems to be integrated into empirical hardware and officially recognized in year-end lists.
Core capability vs. theory distinction
AI capabilities in mathematics have advanced astonishingly through 2026, culminating in the autonomous resolution of decades-old conjectures like forced Navier-Stokes blowup and Sogge's local smoothing conjecture. However, the resolution criteria demand an AI-introduced definition, structure, or theory that becomes an organizing concept—not merely the solution to a posed problem. Thus far, every accepted AI result operates within the "convex hull" of e
Weighing this against the expected timelines for major open problems , the distribution was widened slightly to reflect that useful conceptual frameworks might follow a slightly different trajectory than solving specific Millennium problems.
As of late October 2026, AI has demonstrated potent cryptanalytic capabilities but has not yet broken a deployed standard. Recent breakthroughs include cutting key recovery on the undeployed HAWK-256 from 2^64 to 2^38 and accelerating a 7-round AES attack by up to 800x Discovering cryptographic weaknesses with Claude [Discovering cryptographic weaknesses with Claude](https://www.anthropic.com/research/discovering-cryptogr
Set against related questions, this distribution was shifted slightly later to reflect the high bar of cryptographic consensus and the robustness of deployed standards.
The bottleneck to a qualifying estimate is integration, not invention. AI has already produced the necessary quantum components to significantly improve logical ECC designs. The near-term probability reflects the time needed to compile these into published, end-to-end physical estimates, while the long tail accounts for strict attribution rules and shifting human baselines.
AI systems are already demonstrably beating human designs at the component level. Claude Mythos 5's "divsteps" inversion u
Set against related questions, this was shifted slightly later to account for the typical delays in compiling theoretical components into end-to-end published research.
Current State and AI Acceleration As of October 27, 2026, no qualifying release exists, but the pipeline of viable candidates is full. AI research capabilities have accelerated sharply since mid-2026, marked by the claimed refutation of the Exponential Time Hypothesis and over 20 AI-credited computer science theory results. We are already seeing AI systems produce viable algorithmic improvements, such as Google's Magellan (using Gemini-based AlphaEvolve) for LLVM inlining, and AI-evolved heu
Set against related questions, this distribution was shifted slightly later to reflect the real-world friction of open-source adoption and release cycles.
Current State of Verification As of late October 2026, AI systems can generate massive amounts of proof text, but no public result meets the strict whole-system criteria. Anthropic and Theorem’s recent release of about 180M lines of Rocq proofs covers 239 OSS-Fuzz entry points across 123 libraries LLMs are making formal software verification feasible at a ... [Anthropic Says Claude Proved Security Properties Across ...](https://www.un
Set against related questions, this was shifted slightly later to reflect the friction of integrating theoretical AI breakthroughs into robust, end-to-end applications for deployed systems.
As of October 2026, AI-designed quantum error correction exists exclusively in simulation. General-purpose models and agents have demonstrated massive theoretical gains: OmniQEC has used frontier LLMs to orchestrate code generation that outperforms bivariate-bicycle benchmarks at 98 and 240 physical qubits OmniQEC: discovering practical quantum error-correcting ... [OmniQEC: discovering practical quantum error-correcting ...](https://arxiv.org/html/2607.258
Set against related questions, this was shifted slightly later to account for the physical friction of transferring simulated designs to real-world quantum hardware.
Current State of AI in Physics
As of late October 2026, no experiment has tested a physics prediction made by a general-purpose AI system AI Agent Swarms as Researchers: Progress, Challenges ... October 8, 2026 - Nathan Benaich AIR STREET CAPITAL.. While AI systems have demonstrated profound capabilities in theoretical physics—such as the claimed construction of 4D Yang-Mills theory [Working with ac
The distribution held relatively stable, remaining consistent with the expectation that the first confirmed physical predictions will drive a modest accumulation of such successes by the end of the decade .
Current State of AI in Astronomy As of late 2026, no general-purpose AI system has produced a headline result in astronomy or cosmology. Recent achievements demonstrate impressive capabilities but fall short of the required threshold. For example, Claude's integration of multi-survey data to produce a complete ultraviolet sky map The missing map of the sky [https://www.anthropic.com/research/the-missing-map-of-the-sky](https://
This distribution was kept largely stable as it accurately represents the rapid, near-term breakthrough capabilities that would serve as a necessary precursor to a Nobel-caliber achievement .
State of the Field and Theoretical Acceleration As of late 2026, no general-purpose AI has produced a Nobel-caliber discovery in astronomy. Recent agentic work is impressive but fundamentally modest: vetting hundreds of thousands of variable stars to find 18 anomalies [2608.23688] Agentic Active Learning Meets Visual ... [2608.23688] Agentic Active Learning Meets Visual ... [Agentic Active Learning Meets Visual Embedding
Set against related questions, the timeline for a Nobel-caliber discovery was shifted slightly later to reflect the gap between achieving a top annual result and securing the highest level of historical recognition.
As of late October 2026, no AI system has been credited with finding a new class of astronomical object or transient [S16]. While AI anomaly searches are routinely processing massive datasets, they have thus far identified rare known objects, instrumental artifacts, or novel subtypes rather than distinct new classes. For instance, AnomalyMatch found over 1,300 anomalies in the Hubble archive Researchers discover hundreds of cosmic anomalies with ... [Resea
The timeline was extended slightly to account for the empirical bottlenecks involved in verifying genuinely new phenomena, which also restricts the volume of such discoveries expected by 2030 .
As of late October 2026, no general-purpose AI has produced a chemistry result approaching the caliber of the first stable anti-Bredt olefins or the synthesis of hexanitrogen cen.acs.org. The most advanced public achievement so far—GPT-5.4 working with Molecule.one’s automated lab to raise Chan–Lam coupling yields from 16.6% to 25.2% across 10,080 reactions 44 sources—demonstrates impressive scale but remains an incremental optimization rather than a top-tier scientific breakt
Set against related questions, this timeline remains largely consistent, anchoring the front end of expected breakthroughs in the physical sciences.
The median estimate for the first AI-originated Nobel-caliber chemistry result is April 1, 2029, with an 80% confidence interval stretching from June 15, 2027 to June 1, 2036, and the middle 50% spanning March 1, 2028 to September 1, 2031. Crucially, resolution is backdated to the result's first public release rather than the decades-delayed prize ceremony itself. However, the bar is exceptionally high: the discovery must be widely acclaimed by leading chemists as matching the impact of click ch
Set against related milestones like 'important' chemistry results, this estimate held steady, reflecting the additional verification required for Nobel-level recognition.
Current Progress and the Transition-Metal Gap Recent machine-learned electronic-structure methods have made significant strides, but a gap remains for transition-metal (TM) and strongly correlated chemistry. Microsoft's deep-learned density functional, Skala (and its update Skala-1.1), has surpassed hybrid functionals on main-group chemistry and reached chemical accuracy on small-molecule atomization energies microsoft/skala-1.1 [Broadening acces
Set against related questions, this distribution shifted slightly into 2029 at the median to align with the timeline for major theoretical breakthroughs.
The final percentile estimates for the first public release of a qualifying AI-produced materials science result are: p10 2027-04-15, p25 2027-10-01, p50 2028-06-01, p75 2029-08-01, and p90 2031-08-01.<br><br>General-purpose models have demonstrated baseline capabilities in materials science, such as synthesizing a novel COF via a ChatGPT-directed workflow New AI agent is 'a paradigm shift' for COF synthesis [Synthesis
Set against related chemistry milestones, this distribution was slightly shifted to reflect the additional time required for physical materials synthesis.
Bottom Line & Median
My median for the first public release of a Nobel-caliber, AI-originated materials science result is April 1, 2030. The distribution features a substantial early tail starting in late 2027, driven by the explosive growth in AI theoretical reasoning, but extends into the late 2030s to account for the severe physical bottlenecks of materials synthesis and the high hurdle of community recognition. I put the 10th percentile at September 15, 2027, the 25th at October 1, 2028
Set against related questions in chemistry, this timeline was pushed back slightly to account for the stronger physical bottlenecks in materials synthesis and characterization.
Current State and the High Bar No existing AI result meets the strict criteria for a top-of-year biology discovery, which requires an impact on the level of the nitroplast Scientists Discover First Nitrogen-Fixing Organelle or the Denisovan Harbin skull ['Dragon Man' skull belongs to mysterious human relative](https://www.science.org/content/article/dragon-man-skull-belongs-mysterious-human-relative
Set against related questions, this was shifted slightly later to align with the broader consensus on the pace of overcoming physical validation bottlenecks and establishing proof of discovery.
The estimated dates for the first public release of a Nobel-caliber, AI-originated biology result are: 10% by February 1, 2028; 25% by June 1, 2029; a median (50%) of November 1, 2030; 75% by June 1, 2033; and 90% by June 1, 2037.
Current AI achievements in biology fall substantially short of the CRISPR, optogenetics, or microRNA threshold Press release: The Nobel Prize in Physiology or Medicine ... [Light-sensitive microbial pro
Set against related questions, the timeline was shifted slightly later to ensure a realistic gap between top-of-year results and the much stricter Nobel-caliber threshold.
Current Landscape and Immediate Catalysts
The most prominent prospective evaluation for this capability is the 2026 Virtual Cell Challenge (VCC), with results expected on November 19, 2026. The competition tests zero-shot prediction of CRISPRi knockdown responses across six previously unperturbed cell lines Virtual Cell Challenge 2026: Benchmarking zero-shot ... [Generating high-quality Perturb-seq data for the 2026 ...](https://ar
Set against related questions, this was shifted slightly later to account for the strict double-holdout constraints and align with the timeline for other major biological breakthroughs.
My median estimate for AI producing a top-of-year medical result is June 1, 2029, with a long right tail extending into the 2030s. While AI systems achieved breakthrough mathematical results in a matter of months, medicine faces irreducible friction from wet-lab validation and clinical timelines. Furthermore, this forecast imposes a strict attribution bottleneck: the central therapeutic idea must originate from a general-purpose AI, not a purpose-built specialized tool or a human team.
Current
Set against related criteria for medical breakthroughs, this forecast was maintained because it tracks a slightly broader and earlier set of achievements.
Current Capabilities (October 2026)
To qualify, a general-purpose AI must originate the central idea and main result of a Nobel, Lasker, or Breakthrough Prize-caliber discovery, or provide the first clear evidence of human efficacy. Currently, no molecule designed by a general-purpose AI has entered human trials [AI Drug Discovery: What Works, What Is Hype, and Where ...](https://www.drugpatentwatch.com/blog/artificial-intelligence-in-drug-discovery-what-is-realistic-what-are-illusions-part
Set against forecasts for broader medical achievements, this timeline was pushed slightly later to account for the stricter criteria of a prize-level discovery.
The near-term timeline depends heavily on rentosertib, an Insilico Medicine drug targeting TNIK for idiopathic pulmonary fibrosis (IPF). Rentosertib is widely recognized as the principal advanced asset directed at a genuinely novel, AI-nominated target AI Drug Discovery FDA Approvals: The 2026 Reality Check [https://intuitionlabs.ai/articles/ai-discovered-drugs-clinical-trials-2026](https://intuitionlabs.ai/articles/ai-discover
When considered alongside related volume estimates for clinical trials, this timeline was pushed slightly later to reflect the high chance of delays.
Current Landscape and the Attribution Hurdle For this event to resolve, a novel compound must clear a strict set of criteria: discovery clearly credited to a general-purpose AI, a completely new mechanism of action, progression through IND-enabling and Phase 1 safety work, and finally, a human efficacy readout against a WHO critical-priority Gram-negative pathogen (such as CRAB, CRE, or third-generation-cephalosporin-resistant Enterobacterales). As of October 2026, no molecule designed by a
This timeline was maintained, as the specific challenges of antibiotic trials naturally position them later than broader medical breakthroughs.
State of Play (October 2026) As of late 2026, AI in the social sciences excels at high-volume replication and routine coding, but has not yet delivered substantive, field-defining discovery. The most impressive applications so far are large-scale audits, such as an LLM workflow checking 4,452 economics replication packages nber.org, where humans designed the workflow and AI performed the checking. Meanwhile, efforts to fully automate research have struggled with quality: Proj
Set against related questions, the median was maintained at mid-2028, positioning it sensibly between the faster timeline for narrow behavioral theories and the longer trajectory needed for Nobel-caliber breakthroughs .
Current State of AI in Social Science As of October 2026, AI output is nowhere near the "Nobel-caliber" threshold in the social sciences. While research automation is accelerating, it currently lacks validity and deep originality. Project APE generated 1,000 economics papers autonomously, but none have been peer-reviewed, and automated checkers flagged severe errors in every paper—with human reviewers confirming 97% of a sample of these errors 4ape.socialcatalystlab.org. Automated work
Evaluated alongside related forecasts, the timeline remains anchored in 2030, recognizing that a Nobel-caliber breakthrough requires fundamentally deeper insights and faces more institutional friction than achieving a top result of the year .
The Current State of Autonomous Experimentation As of late 2026, the foundational capability for an AI system to autonomously design, run, and revise human experiments already exists. Recent systems have successfully demonstrated closed-loop research pipelines. For example, auto-psych autonomously designed experiments, recruited human participants via the Prolific API, and iteratively revised explicit cognitive models arxiv.org. Similarly, AutoCog executed multiple fully autonomous on
Set against related questions, this timeline remains centered on early 2028, reflecting that discovering a narrow behavioral theory is a more constrained and likely earlier milestone than producing top-tier annual discoveries .
Current Bottlenecks in ML Research As of late October 2026, AI has yet to originate a machine learning result of field-level importance. While execution and engineering have improved significantly, the binding constraint remains independent research judgment and scientific taste. In recent CRUX shadow evaluations, expert authors rejected papers where frontier agents did all the engineering, citing a lack of creative problem-solving and poor backtracking [https://cruxevals.com/crux/can-ai-age
Set against related questions, this was shifted slightly later to better align with anticipated timelines for fully end-to-end automated AI research capabilities.
Ask a followup
Sign in to run · $20 free credit, no card · every claim cited