AI safety reports need the right model and the right date
By OnTheRice · Sources and limitations · Report a correction
We connect a new Chinese-model disclosure audit with earlier outside evaluations. The useful distinction is between evidence available at launch, evidence published later and tests that match the version being used. A missing developer report cannot stand in for a safety verdict.
Research reference · source time Not applicable: no market-price call · not a current quote or execution price
Source confidenceOriginal reports read; census and model tests not independently reproduced
Published through Sites . The forecast window starts at its frozen preparation time above.
What’s changing?
We distinguish the developer’s public reporting record from outside testing and the evidence available for a buyer’s current model. The new census and the historical evaluations answer different questions.
Key points
- Keep launch-time developer disclosure and later external testing as separate facts.
- Match the exact model, tested tasks and settings before carrying a result into a buying decision.
- Compare like denominators; company counts and release counts cannot form one trend.
What could change this view?
- A missing developer report does not establish absent private testing or a harmful model. Outside reports can be useful but may cover a different version or a narrow use.
Bottom line
We would ask three separate questions before relying on an AI safety report: Is it for this model? Was it available at the time being discussed? Does it cover how we will use the model? Keep developer disclosures and outside tests visible side by side. Late evidence can be valuable, while a missing public report leaves uncertainty. Neither condition alone gives a universal safety verdict.
Its source facts and prices still need independent checks.
How likely is this call? · Three frameworks
This signal has no investible bullish or bearish call. A market-direction percentage does not apply.
Evidence behind this view
Checks whether separate pieces of evidence support the same view.
Four original research or evaluation origins, with different scopes. They do not constitute four independent replications of the disclosure census.
What past price moves show
Compares today’s price moves with recorded past patterns.
Earlier OTR launch-readiness and controls-testing cards were checked. This view adds developer-versus-external evidence and disclosure-time matching, without changing those records.
Evidence against this view
Shows evidence and risks that could challenge this view.
Independent reports demonstrate that developer reporting gaps and outside evaluation can coexist; the external studies themselves contain important coverage limits.
Why the direction changed
Explains recorded changes from expecting a rise to expecting a fall, or the reverse.
New noninvestment view; no earlier asset direction, price, deadline or result is revised.
What people say vs the facts
Compares recorded public discussion with checked market facts.
No public sentiment, adoption rate or population-wide safety claim is made.
How past forecasts performed
Shows recorded forecast results and what they can teach us.
Disclosure shares and benchmark findings are not probabilities of a future incident or asset return.
Sources, original wording and price checks
Author-submitted research. The underlying facts and market prices have not been independently verified.
Forecast prepared
No explicit investible BULLISH or BEARISH call on this card.
What changed
Recorded public historyWe connect a new Chinese-model disclosure audit with earlier outside evaluations. The useful distinction is between evidence available at…
This is the first recorded public view. Further public updates will appear here.
- AI safety reports need the right model and the right date
We connect a new Chinese-model disclosure audit with earlier outside evaluations. The useful distinction is between evidence available at launch, evidence published later and tests that match the version being used. A missing developer report cannot stand in for a safety verdict.
Forecast prepared: 9 Oct, 22:22 SGTPublished by OTR: 9 Oct 2026, 22:55 SGT
Why the view changed
This is the first recorded view. Future source-backed updates will appear here.
Evidence, memory & review 4 linked sources
Directional takeaway: Not applicable: no investment call
OTR reasoning: We distinguish the developer’s public reporting record from outside testing and the evidence available for a buyer’s current model. The new census and the historical evaluations answer different questions.
Counterpoint: A missing developer report does not establish absent private testing or a harmful model. Outside reports can be useful but may cover a different version or a narrow use.
Source-grounded facts (paraphrased)
- Time not supplied: Census reports 31 developer result disclosures split into 9 by launch, 16 later and 6 unresolved in timing or model match. Source record: SEMIANALYSIS-DISCLOSURE-20261008
- Time not supplied: Industry review uses developer-level coverage, with five of ten disclosing alongside at least one release in its July 2025–June 2026 period. Source record: CONCORDIA-CHINA-SAFETY-202607
- Time not supplied: Independent Kimi K2.5 assessment explicitly leaves vision, agent swarm and most languages untested. Source record: INDEPENDENT-KIMI-K25-20260403
- Time not supplied: CAISI assessment concerns Kimi K2 Thinking, not Kimi K2.5. Source record: NIST-CAISI-K2THINKING-20251212
Source trail
- Beijing Will Not Pace the Frontier: China’s Speed-First AI Safety Regime ↗SemiAnalysis / Mark Chen, Doug and Dylan Patel8 October 2026
- State of AI Safety in China: July 2026 executive summary ↗Concordia AIJuly 2026 report
- An Independent Safety Evaluation of Kimi K2.5 ↗arXiv / Zheng-Xin Yong and coauthors2026-04-03T15:45:35Z · published (UTC)
- CAISI Evaluation of Kimi K2 Thinking ↗NIST / Center for AI Standards and Innovation12 December 2025; updated 9 January 2026
Review status: PASSED
No score or probability is presented.