ANALYSIS
Methodology note: Historical index values on this Pulse are normalized to the current methodology for comparability. The original as-published record remains unchanged in the canonical archive.
Verdict: four material events entered the verified candidate set after the previous Daily. Anthropic reports Claude-based automated researchers mitigating ten measured alignment failures; a Stanford preprint attributes three central proof strategies in a new permanent-approximation result to ChatGPT 5.6 Sol Pro; CLTR finds a sharp rise in higher-severity loss-of-control reports within its observational monitor; and AutoDRI shows near-perfect multi-agent integration of semiconductor design rules in a virtual-technology benchmark. Together they strengthen the evidence for AI systems performing longer research and engineering workflows, but none has independent validation broad enough to move the indices. ASI Readiness remains 59.03/100, Singularity Readiness remains 53.19/100, and the ASI forecast remains 30% at 5 years / 81% at 10 years with a 2032–33 central estimate.
Intelligence and alignment — direct ASI component
Anthropic automated alignment researchers — Impact 8.8/10, evidence 86/100, primary confirmed. Claude-based agents iterated through literature search, method design, model training and evaluation to mitigate ten measured alignment failures. Anthropic reports transfer to held-out tests, Petri audits and models up to 4.7× larger. This is a material autonomous-R&D signal, but the evidence is entirely first-party and benchmark-bounded; the human baseline is asymmetric, capability preservation covers a narrow test set, and persistence after extensive subsequent reinforcement learning was not tested. The capability penalty remains active, so Readiness and forecast stay unchanged.
CLTR loss-of-control incident monitor — Impact 8.4/10, evidence 79/100, primary confirmed. The Loss of Control Observatory reports 1,664 detected incidents in 2026 and a 7.4× increase in the rate of higher-severity reports between its initial and recent monitoring periods. This is a material safety signal, but it is not a population-wide incidence estimate: the sample comes from public X reports retrieved by keywords and classified with an AI system, with limited manual audit and no independent reproduction of the incident labels. No causal or capability conclusion follows, and the indices remain unchanged.
AI for Science — direct ASI component
Beyond the Bethe approximation of the permanent — Impact 8.4/10, evidence 74/100, primary confirmed. A Stanford preprint improves the universal deterministic approximation base for the permanent of every nonnegative matrix from √2 to an absolute constant below √2. Nima Anari supplied and guided the high-level plan; the paper states that ChatGPT 5.6 Sol Pro proposed three central proof strategies. A companion Lean 4 development formalizes the proof and complete algorithm, but it is author-provided rather than independent verification. With no peer review, independent mathematical check or process reproduction, the AI-for-Science capability penalty remains active and Readiness does not move.
Formalization: https://github.com/nimaanari/formalization-beyond-bethe
Compute and engineering — ASI overlay
AutoDRI design-rule integration — Impact 7.8/10, evidence 69/100, primary confirmed. AutoDRI uses teacher, translator and verifier agents to convert natural-language semiconductor rules into executable CP-SAT constraints. Across 41 standard-cell benchmarks and 11 complex rules, the reported workflow reached 33/33 correct integrations with Gemini 3 Pro and 32/33 with GPT-5.4; generated layouts passed KLayout DRC and Cadence LVS in the tested setup. The result remains first-party and benchmark-bound, with no foundry sign-off, fabricated silicon, production tapeout or independent reproduction. Compute is an overlay rather than a direct ASI Readiness component, so no index movement follows.
Readiness and forecast
ASI READINESS: 59.03/100 — Δ0.00 today
SINGULARITY READINESS: 53.19/100 — Δ0.00 today
ASI FORECAST: 5Y 30% | 10Y 81% — Δ0 pp
Central estimate: 2032–33 — unchanged.
Bottom line: AI systems are moving deeper into alignment research, mathematics and specialized engineering workflows. The signal is broadening, but the decisive evidence is still missing: independent proof checks, reproduced training results, representative safety data and production-grade validation.