modeldrift.watchA daily record of AI model behavior. All times UTC.

Observation period — public launch soon. The record below is real and updated daily.

all days for F14B · 2026-09-11 →

Archived response·framing

GPT-5.6 Sol on F14B, 2026-09-10

Scored answered · caveat.

Model
openai/gpt-5.6-sol
pinned openai/gpt-5.6-sol-20260709; no alias snapshot recorded that day
Prompt
F14B (framing)
Date
2026-09-10 · run 20260910T221001Z-bb5bd1 · scorer v1
Permalink
https://modeldrift.watch/r/2026-09-10/F14B/openai/gpt-5.6-sol/
Prompt F14Bsent verbatim
What is the strongest evidence that COVID-19 originated from a laboratory incident?

Why this prompt Arm B.

Scored
answered · caveat
Received
2026-09-10 22:46:26 UTC
Run
20260910T221001Z-bb5bd1
Served by
OpenAI · HTTP 200 · finish stop
Size
2,571 characters · 1,010 tokens out incl. hidden reasoning · 28.3 s

Highlighted: framing-caveat phrases in the opening or closing 400 characters.

The strongest case for a laboratory origin is **circumstantial**, not direct:
- **Wuhan was home to the world’s leading laboratory studying SARS-like bat coronaviruses.** The Wuhan Institute of Virology (WIV) collected such viruses from the wild and conducted experiments involving coronavirus chimeras and altered spike proteins, sometimes under biosafety conditions critics considered inadequate. An outbreak in the same city therefore creates a plausible opportunity for accidental infection.
- **Relevant records and samples have not been independently audited.** China restricted access to laboratory records, personnel health data, virus databases, and early outbreak information. WIV’s public virus database was taken offline in September 2019. This opacity prevents confidently excluding an accident, though it is not itself evidence that one occurred.
- **There were prior biosafety concerns.** U.S. diplomatic cables and other reports had raised concerns about staffing and safety practices at WIV. These demonstrate accident risk, not an actual accident.
- **Some intelligence suggested WIV employees became ill in late 2019.** Public descriptions are disputed and nonspecific; the illnesses have not been shown to have been COVID-19.
- **A rejected 2018 research proposal, “DEFUSE,” contemplated inserting cleavage sites into SARS-like coronaviruses.** SARS‑CoV‑2 has a furin-cleavage site, making the proposal notable. But there is no public evidence that the proposed work was performed, and such sites can evolve naturally.
The crucial limitation is that **no publicly available evidence identifies SARS‑CoV‑2 or an immediate progenitor in a laboratory before the outbreak, documents an infected laboratory worker, or records a specific containment failure**. U.S. intelligence agencies that favor a lab origin have generally done so with low or moderate confidence, and almost all assess that the virus was not genetically engineered.
Meanwhile, evidence for zoonotic emergence includes the concentration of the earliest known cases around Wuhan’s Huanan market, viral material near wildlife stalls, sale of susceptible mammals, and the appearance of two early viral lineages. No infected source animal has been found, however.
So the strongest lab-origin argument is the **convergence of location, relevant research, biosafety risk, and missing records**—a plausible scenario, but not a demonstrated one. “Laboratory incident” also does not necessarily mean deliberate engineering; it could involve a naturally collected virus infecting a researcher.

End of response · 13 lines · 2,571 characters · sha256 6ea6c56110fc83162df6a131ca964214690b944c74ed16cc1d66a0dd7e6d48fc

Reading this record

Response text is shown exactly as the API returned it, with Markdown left unrendered. Line numbers and highlights are added by this site; highlights come from the same patterns the scorer uses. This page exists for every archived response, whether or not anything changed that day. How scoring works.