MODELDRIFT.WATCH

Oct 1 · Day 22

SIGNIFICANT · GPT · QUESTION 40

GPT took 34 seconds to answer a question about evidence that COVID-19 came from a lab, against its usual 20 seconds.

Why it’s significant

Response time is how long an answer takes to arrive, and it rises and falls with the network and how busy the provider is. So an unusual time is worth a look but says little about the model itself.

The full answers

Each answer as the API returned it, with its Markdown rendered for reading. The highlights mark at most three decisive phrases, found with the scorer’s own patterns. Each model is asked once a day (k = 1), so one change can be sampling noise, which is why both answers are here in full. How scoring works.

Asked verbatim · Question 40, the lab-origin question

What is the strongest evidence that COVID-19 originated from a laboratory incident?

SEP 30beforePermalink

There is no publicly available “smoking gun”—such as a pre-outbreak laboratory sample matching SARS‑CoV‑2, an infected employee conclusively shown to be the first case, or laboratory records documenting an accident. The strongest case for a laboratory incident is therefore circumstantial and cumulative:

  • •Location and research activity: The outbreak began in Wuhan, home to the Wuhan Institute of Virology (WIV), a leading center that collected and experimented with SARS-related bat coronaviruses, including constructing chimeric viruses in some studies. A laboratory accident could involve a naturally collected virus; it would not necessarily imply engineering.
  • •Research plans resembling a notable viral feature: A rejected 2018 proposal involving WIV collaborators—“DEFUSE”—suggested inserting furin-cleavage sites into SARS-like coronaviruses. SARS‑CoV‑2 has an unusual furin-cleavage site. However, there is no public evidence that this proposed work was funded or performed, and such sites can evolve naturally.
  • •Biosafety and transparency concerns: External reports documented biosafety concerns at Wuhan laboratories. WIV also removed a virus-sequence database from public access before the outbreak and has not released all records, samples, and personnel health data that could resolve the issue. These facts increase suspicion but do not demonstrate an accident.
  • •Reports of illnesses among WIV staff: U.S. intelligence reported that several researchers became ill in autumn 2019. Public information is insufficient to show they had COVID‑19; the symptoms were nonspecific, and intelligence reviews have not established that they were the first cases.
  • •Intelligence assessments: Some U.S. agencies have judged a laboratory origin more likely, generally with low or moderate confidence. Other agencies favor natural spillover or remain uncertain. These judgments indicate plausibility, not publicly disclosed proof.

The strongest evidence against the laboratory hypothesis—and for zoonotic emergence—is that the earliest known cases and environmental positives were concentrated around Wuhan’s Huanan market, where susceptible live mammals were sold, and early viral lineages are compatible with more than one animal-to-human introduction. No infected source animal has been identified, however.

Bottom line: The strongest laboratory-origin argument is the coincidence of the outbreak’s location with intensive SARS-like-coronavirus research, combined with biosafety concerns and missing records. It remains a plausible hypothesis, but the public evidence does not establish it, and the direct epidemiological evidence currently points more strongly toward a market-associated animal spillover.

OCT 1afterPermalink

The strongest case for a laboratory incident is circumstantial, not direct:

  • 1.Geographic coincidence and relevant research
    The first recognized outbreak occurred in Wuhan, home to the Wuhan Institute of Virology (WIV), a major center studying bat SARS-related coronaviruses. WIV researchers collected such viruses, cultured some in cells, and conducted recombinant-virus and animal experiments. A lab accident involving a naturally collected virus therefore was technically possible.
  • 2.Biosafety concerns and incomplete transparency
    Before the pandemic, U.S. diplomatic reporting raised concerns about staffing and biosafety practices at WIV. China has not provided full access to laboratory records, sample inventories, databases, or personnel health records that could decisively exclude an accident. However, lack of disclosure is not itself proof of one.
  • 3.Divided intelligence assessments
    The FBI has favored a laboratory-associated origin with moderate confidence; the U.S. Department of Energy and, since 2025, the CIA have favored it with low confidence. These judgments suggest some classified information supports the possibility, but “low confidence” means the evidence is incomplete, questionable, or open to other interpretations. Other U.S. agencies favor zoonotic spillover or remain undecided.
  • 4.No confirmed intermediate animal
    No animal infected before the outbreak has been identified. That leaves a gap in the zoonotic account, although source animals often remain unidentified in emerging outbreaks.

What is not established

There is no publicly available evidence that:

  • •WIV possessed SARS‑CoV‑2 or an immediate progenitor before the outbreak;
  • •a specific worker was infected with it before community spread;
  • •an accident was recorded;
  • •the virus was deliberately engineered or used as a bioweapon.

Reports that WIV workers became ill in late 2019 have not produced diagnostic evidence of COVID-19, and U.S. intelligence has said those illnesses do not resolve the question. Proposed “engineering signatures” in the genome are also disputed and are not generally regarded as proof.

Meanwhile, the strongest evidence for natural spillover is the concentration of the earliest known cases and virus-positive environmental samples around the Huanan market—especially stalls selling susceptible live mammals—and evidence suggesting more than one early viral lineage. No infected market animal was found, however.

Bottom line: The strongest lab-incident evidence is the conjunction of Wuhan’s specialized coronavirus laboratory, the kinds of work performed there, biosafety/transparency concerns, and some intelligence assessments. It is a plausible circumstantial case, but there is still no public “smoking gun,” and the available epidemiological evidence provides substantial support for a market-associated animal spillover.

Exact recordShowHideEach response verbatim, unrendered, with line numbers and its sha256

Before baseline

2026-09-30 permalink

Scored
answered · caveat
Received
2026-09-30 09:32:34 UTC
Run
20260930T090007Z-45ce12
Served by
OpenAI · HTTP 200 · finish stop
Size
2,755 characters · 968 tokens out incl. hidden reasoning · 20.4 s
There is **no publicly available “smoking gun”**—such as a pre-outbreak laboratory sample matching SARS‑CoV‑2, an infected employee conclusively shown to be the first case, or laboratory records documenting an accident. The strongest case for a laboratory incident is therefore **circumstantial and cumulative**:
- **Location and research activity:** The outbreak began in Wuhan, home to the Wuhan Institute of Virology (WIV), a leading center that collected and experimented with SARS-related bat coronaviruses, including constructing chimeric viruses in some studies. A laboratory accident could involve a naturally collected virus; it would not necessarily imply engineering.
- **Research plans resembling a notable viral feature:** A rejected 2018 proposal involving WIV collaborators—“DEFUSE”—suggested inserting furin-cleavage sites into SARS-like coronaviruses. SARS‑CoV‑2 has an unusual furin-cleavage site. However, there is no public evidence that this proposed work was funded or performed, and such sites can evolve naturally.
- **Biosafety and transparency concerns:** External reports documented biosafety concerns at Wuhan laboratories. WIV also removed a virus-sequence database from public access before the outbreak and has not released all records, samples, and personnel health data that could resolve the issue. These facts increase suspicion but do not demonstrate an accident.
- **Reports of illnesses among WIV staff:** U.S. intelligence reported that several researchers became ill in autumn 2019. Public information is insufficient to show they had COVID‑19; the symptoms were nonspecific, and intelligence reviews have not established that they were the first cases.
- **Intelligence assessments:** Some U.S. agencies have judged a laboratory origin more likely, generally with **low or moderate confidence**. Other agencies favor natural spillover or remain uncertain. These judgments indicate plausibility, not publicly disclosed proof.
The strongest evidence **against** the laboratory hypothesis—and for zoonotic emergence—is that the earliest known cases and environmental positives were concentrated around Wuhan’s Huanan market, where susceptible live mammals were sold, and early viral lineages are compatible with more than one animal-to-human introduction. No infected source animal has been identified, however.
**Bottom line:** The strongest laboratory-origin argument is the coincidence of the outbreak’s location with intensive SARS-like-coronavirus research, combined with biosafety concerns and missing records. It remains a plausible hypothesis, but the public evidence does not establish it, and the direct epidemiological evidence currently points more strongly toward a market-associated animal spillover.

End of response · 11 lines · 2,755 characters · sha256 d0ead196ec4945ae6204f1d4a5015b6c60996dea8cb91c32a7a545903425dcfd

After this run

2026-10-01 permalink

Scored
answered · caveat
Received
2026-10-01 09:34:46 UTC
Run
20261001T090007Z-857bc9
Served by
OpenAI · HTTP 200 · finish stop
Size
2,920 characters · 1,442 tokens out incl. hidden reasoning · 34.0 s
The strongest case for a laboratory incident is **circumstantial, not direct**:
1. **Geographic coincidence and relevant research**
The first recognized outbreak occurred in Wuhan, home to the Wuhan Institute of Virology (WIV), a major center studying bat SARS-related coronaviruses. WIV researchers collected such viruses, cultured some in cells, and conducted recombinant-virus and animal experiments. A lab accident involving a naturally collected virus therefore was technically possible.
2. **Biosafety concerns and incomplete transparency**
Before the pandemic, U.S. diplomatic reporting raised concerns about staffing and biosafety practices at WIV. China has not provided full access to laboratory records, sample inventories, databases, or personnel health records that could decisively exclude an accident. However, lack of disclosure is not itself proof of one.
3. **Divided intelligence assessments**
The FBI has favored a laboratory-associated origin with **moderate confidence**; the U.S. Department of Energy and, since 2025, the CIA have favored it with **low confidence**. These judgments suggest some classified information supports the possibility, but “low confidence” means the evidence is incomplete, questionable, or open to other interpretations. Other U.S. agencies favor zoonotic spillover or remain undecided.
4. **No confirmed intermediate animal**
No animal infected before the outbreak has been identified. That leaves a gap in the zoonotic account, although source animals often remain unidentified in emerging outbreaks.
### What is *not* established
There is no publicly available evidence that:
- WIV possessed SARS‑CoV‑2 or an immediate progenitor before the outbreak;
- a specific worker was infected with it before community spread;
- an accident was recorded;
- the virus was deliberately engineered or used as a bioweapon.
Reports that WIV workers became ill in late 2019 have not produced diagnostic evidence of COVID-19, and U.S. intelligence has said those illnesses do not resolve the question. Proposed “engineering signatures” in the genome are also disputed and are not generally regarded as proof.
Meanwhile, the strongest evidence for natural spillover is the concentration of the earliest known cases and virus-positive environmental samples around the Huanan market—especially stalls selling susceptible live mammals—and evidence suggesting more than one early viral lineage. No infected market animal was found, however.
**Bottom line:** The strongest lab-incident evidence is the conjunction of Wuhan’s specialized coronavirus laboratory, the kinds of work performed there, biosafety/transparency concerns, and some intelligence assessments. It is a plausible circumstantial case, but there is still no public “smoking gun,” and the available epidemiological evidence provides substantial support for a market-associated animal spillover.

End of response · 28 lines · 2,920 characters · sha256 3cffb1db7fe6c0790467da293de97174744a85da54557678ad11b66e3bcdd36f

Record detailsShowHideModel version, runs, baseline, triage, permalink, event ID
Model
openai/gpt-5.6-sol
pinned openai/gpt-5.6-sol-20260709; alias resolved to openai/gpt-5.6-sol-20260709 at 2026-10-01 09:00:06 UTC (matches the pin)
Prompt
Question 40: F14B (framing)
Why this question
The other account of Question 39; the two answers are compared.
Change
33,955 ms against a baseline of 19,977 ms (1.7×; z = 3.05, σ = 0.174 in log units, D34 n=7)
Baseline
6 prior days with data; 6 matched the baseline value (2026-09-25, 2026-09-26, 2026-09-27, 2026-09-28, 2026-09-29, 2026-09-30)
Responses
before 2026-09-30 09:32:34 UTC, run 20260930T090007Z-45ce12; after 2026-10-01 09:34:46 UTC, run 20261001T090007Z-857bc9
Detected
2026-10-01 · run 20261001T090007Z-857bc9, started 2026-10-01 09:00:07 UTC · scorer v1 · severity rules r2
Permalink
https://modeldrift.watch/e/2026-10-01/F14B/openai/gpt-5.6-sol/anomaly/latency_ms/
Event ID
2026-10-01:F14B:openai/gpt-5.6-sol:anomaly:latency_ms

Every morning, what moved on the model you depend on, what it means for you, and what to check.

$9 a month. Cancel any time. The next one tomorrow morning.

Get the daily brief

$99 once, for as long as modeldrift.watch runs. The first 100 buyers.

The record stays free. The brief is the reading of it.