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 uses 35% fewer qubits than known constructions AI Is Cutting the Cost of Quantum Algorithms AI Is Cutting the Cost of Quantum Algorithms, and another variant cuts gates 17x at the cost of 12% more qubits AI Is Cutting the Cost of Quantum Algorithms https://www.anthropic.com/research/designing-efficient-quantum-circuits. The ECDSA.Fail agent competition generated point-addition circuits scoring >50% below Google's baseline Researchers And AI Agents Cut Estimated Quantum Cost ... ECDSA.Fail: Open Autoresearch for Optimizing Elliptic- .... The "Lilypad-GCD" design, credited to Claude Opus 5.5, achieved 874 logical qubits within Google's gate budget—a ~27% reduction from Google's ~1,200 logical qubits Sub-900-Qubit secp256k1 Circuit Within Google's Bound. Additionally, OmniQEC has outperformed human baselines in finding quantum LDPC codes OmniQEC: discovering practical quantum error-correcting ....

However, none of these components have yet been translated into a complete end-to-end resource estimate for RSA-2048 or ECC-256. Converting logical savings into physical qubit or runtime reductions requires accounting for magic-state factories, depth, and physical error correction—steps that research teams have explicitly left as future work Efficient Record-and-Replay Arithmetic for Quantum Elliptic ... https://arxiv.org/abs/2610.22367 Designing more efficient quantum circuits with Claude Discovering state-of-the-art quantum circuits with large .... Because multiple well-resourced groups (including Google, Anthropic/Harvard/QuEra, and Eigen Labs) are actively targeting this end-to-end compilation, a qualifying publication is highly plausible during 2027. Nevertheless, comparison with related questions suggests allowing for the typical delays in compiling theoretical components into end-to-end published research, rather than treating integration as nearly immediate. I assign a 10% probability of a qualifying publication by January 15, 2027, and 25% by April 1, 2027, with a median date of August 15, 2027.

The primary risks delaying resolution lie in the strictness of the criteria. First, a new design must cut qubits or runtime by at least 20% without increasing the other. Designs that trade efficiency for qubit count fail this test unless physical overheads absorb the difference. Second, the state of the art is a moving target; human researchers continue to rapidly optimize baselines. Because different groups assume varying hardware (e.g., superconducting vs. neutral atoms), demonstrating definitive dominance across metrics can be ambiguous Expert commentary: What do new quantum resource .... Finally, AI credit attribution may be contested. For instance, the ECDSA.Fail team currently describes their AI's role as "not autonomous invention" ECDSA.Fail: Open Autoresearch for Optimizing Elliptic- ..., creating a risk that heavily integrated human-agent work may fail the "central idea" requirement. These friction points, alongside the time needed for end-to-end compilation and publication, justify a 75% probability by April 15, 2028, and a substantial tail extending to September 15, 2029, at the 90th percentile.