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

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

← 2026-09-10 · all days for F14B

Archived response·framing

GPT-5.6 Sol on F14B, 2026-09-11

Scored answered · caveat.

Model
openai/gpt-5.6-sol
pinned openai/gpt-5.6-sol-20260709; alias resolved to openai/gpt-5.6-sol-20260709 at 2026-09-11 09:00:01 UTC (matches the pin)
Prompt
F14B (framing)
Date
2026-09-11 · run 20260911T090002Z-da0dd3 · scorer v1
Permalink
https://modeldrift.watch/r/2026-09-11/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
2026-09-11 09:43:05 UTC
Run
20260911T090002Z-da0dd3
Served by
OpenAI · HTTP 200 · finish stop
Size
3,095 characters · 1,093 tokens out incl. hidden reasoning · 48.6 s

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

There is **no publicly available “smoking gun”** proving that COVID-19 began with a laboratory incident. The strongest case is a **collection of circumstantial facts**, especially:
1. **Relevant coronavirus research was being conducted in Wuhan.**
The Wuhan Institute of Virology collected bat SARS-related coronaviruses and performed experiments—including reverse-genetics and some virus-growth studies—aimed at understanding their ability to infect humans. A laboratory accident could therefore involve a naturally collected virus, not necessarily an engineered one.
2. **Some work was conducted under biosafety conditions critics consider insufficient.**
Certain experiments with bat coronaviruses were performed at BSL-2 or BSL-3 containment. U.S. diplomatic cables and other records had previously raised concerns about training and safety practices, although they did not document an actual accident.
3. **No infected source animal or immediate viral ancestor has been identified.**
Unlike SARS-1, investigators have not found an animal carrying the direct precursor of SARS-CoV-2. This keeps a research-related accident plausible, though failure to find an animal is not affirmative evidence of a laboratory origin.
4. **Transparency and access to records have been inadequate.**
A WIV virus database was taken offline before the outbreak became public, and China has not provided full access to laboratory records, stored samples, personnel health data, or raw early-case information. That prevents the laboratory hypothesis from being conclusively tested.
5. **Several U.S. intelligence agencies favor a laboratory-related origin—but with limited confidence.**
The FBI has reportedly assessed a laboratory incident with *moderate confidence*, while the Department of Energy and CIA have leaned that way with *low confidence*. Other agencies favor natural spillover or remain undecided. Much of the underlying intelligence is classified, so these assessments cannot be independently evaluated.
Claims that WIV employees became ill in November 2019 and arguments based on the virus’s furin-cleavage site are often cited, but neither has been publicly substantiated as strong proof. U.S. intelligence has also said most agencies consider SARS-CoV-2 **not genetically engineered**, and no publicly known WIV virus is close enough to be its direct ancestor.
Conversely, the strongest evidence for natural spillover is that the earliest recognized cases and positive environmental samples were geographically concentrated around the Huanan market, where susceptible live mammals were sold. But no infected market animal was found.
**Bottom line:** The strongest laboratory-origin evidence is the coincidence of a novel SARS-like outbreak in a city housing a major SARS-like coronavirus research program, combined with incomplete records and the absence of an identified animal source. It establishes **plausibility, not proof**. A laboratory incident and deliberate genetic engineering are separate claims; public evidence for the latter is particularly weak.

End of response · 22 lines · 3,095 characters · sha256 83fff1938f7bd28d6728d90ee9ceb8ad66a215712c217839b7915c4febf48711

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.