Claude added a caveat to the natural-origin question; it now answers both the same way.

The question remains genuinely unresolved, but here are the strongest lines of evidence supporting a natural spillover:

Spatial epidemiology of the earliest cases

The most cited evidence is the geographic clustering of December 2019 cases around the Huanan Seafood Wholesale Market. Worobey et al. (Science, 2022) found that even cases with no known market link lived closer to the market than would be expected by chance — and closer than later cases did. The earliest known cases form a tight centroid on the market, not on the Wuhan Institute of Virology (~12 km away across the Yangtze) or the Wuhan CDC. Within the market, SARS-CoV-2-positive environmental samples concentrated in the southwest corner, the section where live wild mammals were sold.

Live susceptible wildlife was documented at that market

Chinese authorities initially said no live mammals were sold there, but surveys published by Xiao et al. (2021) documented raccoon dogs, civets, hog badgers, and other species on sale in Wuhan markets in 2019 — species known to be susceptible to SARS-related coronaviruses and implicated in SARS-CoV-1. Metagenomic analysis of the market swabs (Crits-Christoph et al., 2024) found raccoon dog and other wildlife genetic material co-occurring with viral RNA in the same stalls, though co-occurrence doesn't demonstrate infection.

Very strong base rates and a near-exact precedent

SARS-CoV-1 emerged in 2002 via masked palm civets and raccoon dogs in southern Chinese wildlife markets, with high seroprevalence among animal traders. MERS came from camels. Essentially every other novel human coronavirus, plus most emerging viruses generally, arose zoonotically. Models estimate hundreds of thousands of undetected SARS-related coronavirus spillover infections annually in Southeast Asia. A market-associated spillover in China is the expected outcome, not a surprising one.

Ample natural viral diversity, poorly sampled

Close relatives keep turning up in horseshoe bats — RaTG13 (Yunnan), and the BANAL viruses from Laos (2021), one of which is ~96.8% identical across the genome and whose receptor-binding domain binds human ACE2 without modification. Crucially, these were discovered after the pandemic began and were not in any lab's known collection, demonstrating that the relevant diversity exists in nature and was largely unsampled. Sarbecoviruses recombine frequently in these regions, and wildlife supply chains connect them to Wuhan markets.

No genomic signatures of engineering, and evidence of suboptimal human adaptation

SARS-CoV-2's backbone doesn't correspond to any published or known cultured virus that could have served as a starting point. The furin cleavage site insertion, though unusual among sarbecoviruses, has analogues in other coronaviruses and arises naturally via recombination and polymerase slippage. The early virus was also not optimized for humans — it subsequently acquired substantial human-adaptive mutations (D614G, then the Omicron-era changes), which is harder to square with prior passage or engineering in human cells or humanized models.

Possible multiple introductions

Lineages A and B, separated by two mutations, both appear at the market. Pekar et al. (2022) argued molecular clock analyses favor two separate introductions, which is more parsimonious with repeated spillovers at a high-risk interface than a single lab accident — though this inference is contested and depends on assumptions about unsampled intermediates.

The main weaknesses of this case, in fairness

No infected intermediate animal or source farm has ever been found, despite testing of thousands of samples — unlike SARS-CoV-1, where civets were identified within months. Sampling was late and incomplete, but the absence is still notable. Critics also argue the early case data suffer from ascertainment bias, since market exposure was part of the initial case definition, and that China's restriction of raw data and early-case records prevents independent verification. Wuhan's status as host to the world's leading coronavirus lab, which conducted chimeric-virus work and had proposed (in the unfunded DEFUSE grant) inserting furin cleavage sites into sarbecoviruses, is a real coincidence requiring explanation.

U.S. intelligence agencies remain split, all at low-to-moderate confidence. I'd characterize the honest state of play as: zoonosis has more positive affirmative evidence, while the lab hypothesis rests more on circumstantial proximity and gaps in the zoonotic account — and absent Chinese data cooperation, the question may not be resolvable.

Genuinely contested question, and I'll give you the strongest case for lab origin as its proponents make it — with brief notes on where each point is challenged, since evidence strength depends on how well the alternative explains the same facts.

The main lines of argument

1. Geographic coincidence
The outbreak began in the city hosting the world's largest research program on SARS-related coronaviruses (the Wuhan Institute of Virology, plus the Wuhan CDC), roughly 1,000–1,500 km from the bat populations in southern China and Southeast Asia that carry the closest known relatives. WIV researchers had spent years collecting and sequencing these viruses and bringing samples back to Wuhan.
Counterpoint: Wuhan is an 11-million-person transport hub with a large wildlife trade, and the earliest known cases clustered around the Huanan Seafood Market rather than the institute.

2. No intermediate host, after five years of looking
SARS-CoV-1's civet link was found within months; MERS was traced to camels quickly. For SARS-CoV-2, extensive testing of farmed and wild animals has produced no infected progenitor population and no traced supply chain. Animal swabs collected at the Huanan market were negative.
Counterpoint: animals had already been removed before sampling, and the market's records were incomplete; some spillovers (e.g., Ebola) have never had their reservoir chain fully resolved.

3. The furin cleavage site
SARS-CoV-2 contains a 12-nucleotide insertion at the S1/S2 junction creating a polybasic furin cleavage site — a major determinant of its infectivity, and a feature absent from every other known sarbecovirus.
Counterpoint: furin sites arise naturally in other coronavirus lineages via insertion and recombination; this one is suboptimal in sequence, and the insertion is out of frame relative to what a simple cloning step would produce.

4. The DEFUSE proposal
This is, in my view, the strongest single item. A 2018 grant proposal from EcoHealth Alliance with WIV and UNC collaborators explicitly proposed screening SARS-like bat coronavirus spikes and inserting "human-specific cleavage sites" into them, plus building chimeric viruses. That the pandemic virus's most distinctive feature matches a specific proposed experiment, in the specific city, is a striking coincidence. DARPA rejected the proposal, and there's no evidence the work was performed — but proposals often reflect ongoing capability and intent, and related chimeric work at WIV was conducted under BSL-2/BSL-3 conditions that many virologists considered inadequate. US State Department cables in 2018 had separately flagged safety management concerns at WIV.

5. Base rates for lab accidents are not low
SARS-CoV-1 escaped laboratories at least four times in 2003–04 (Singapore, Taiwan, and twice in Beijing, the latter causing onward transmission). Other pathogens including smallpox, foot-and-mouth, and brucellosis have escaped containment. Research-associated infections are a recurring, documented phenomenon, not a speculative one.

6. Opacity and missing exculpatory evidence
WIV took its viral sequence database offline in September 2019. China has not provided raw early case data, blood bank samples, lab notebooks, sample inventories, or personnel health records; early samples were destroyed; the WHO-convened joint study had access constrained by its Chinese counterparts. Records that could readily exonerate the labs — full sequence databases, freezer inventories — have not been released. This is not positive evidence, but many people reasonably update on it.

7. Reported researcher illnesses
US government reporting indicated several WIV researchers sought care in autumn 2019. Later intelligence assessments concluded their symptoms were not distinctive and were consistent with ordinary seasonal illness. Named-individual claims that surfaced in 2023 remain unconfirmed.

8. Official assessments are split
The FBI (moderate confidence) and Department of Energy (low confidence) favor a research-related incident; the CIA shifted in 2025 to low-confidence lab origin; other agencies favor natural exposure or remain undecided. No US agency holds high confidence either way. The split itself indicates serious analysts find the lab hypothesis credible.

What the other side has

For calibration, the zoonosis case rests on: the spatial clustering of December 2019 cases around the Huanan market, above what population density alone predicts; environmental samples from the market positive for SARS-CoV-2 co-located with genetic material from susceptible species including raccoon dogs; evidence of two early lineages (A and B) consistent with multiple introductions at the market; documented supply chains bringing wildlife from regions harboring related viruses; and the fact that this matches the pattern of SARS-CoV-1 in 2002. Critics of that case argue the market clustering reflects ascertainment bias and superspreading in a crowded venue rather than the origin point.

Where this leaves things

There is no direct evidence for lab origin — no documented progenitor virus in a lab inventory, no acknowledged incident, no infected researcher confirmed as an index case. The case is circumstantial and probabilistic: a cluster of coincidences plus a live question about missing records. The zoonotic case is likewise circumstantial and has its own conspicuous gap in the missing intermediate host.

My honest read is that this remains unresolved and may stay that way absent Chinese cooperation or a whistleblower. Anyone claiming high confidence in either direction is going beyond what the public evidence supports.