Claude went back to its usual answers on the COVID-19 origin questions.

The question remains genuinely unresolved, but here is the strongest evidence cited for the zoonotic hypothesis, roughly in order of how much weight it carries in the scientific debate:

1. Spatial clustering of the earliest known cases around the Huanan market

Worobey et al. (Science, 2022) mapped the ~170 December 2019 cases in the WHO-China joint mission data. Cases with no known market link still lived significantly closer to the Huanan market than expected by chance, and closer than later cases did. In an epidemic seeded elsewhere and merely amplified at a market, you would expect market-linked cases to cluster there but unlinked cases to be distributed around the source. This is arguably the single strongest quantitative argument.

2. Environmental positives concentrated in the wildlife-selling section

Of the positive environmental samples the China CDC collected in January 2020, the highest density was in the market's southwest corner — the area where live mammals were sold. Crits-Christoph et al. (Cell, 2024) reanalyzed the raw metagenomic data and found genetic material from raccoon dogs, civets, and bamboo rats in the same stalls and sometimes the same swabs as SARS-CoV-2. This places susceptible host species at the exact location of the earliest viral signal (though co-location is not proof those animals were infected).

3. Documented sale of susceptible species at that market

Despite initial denials, photographic and market-survey evidence (Xiao et al., 2021) confirmed live raccoon dogs, civets, and badgers were sold at Huanan. Raccoon dogs are efficiently infected by SARS-CoV-2 and were implicated in SARS-CoV-1.

4. Strong historical base rate

SARS-CoV-1 emerged through the wildlife trade with market animals as intermediates; MERS via camels. Serological surveys near bat caves in Yunnan and Southeast Asia show repeated, ongoing human exposure to bat sarbecoviruses. Coronavirus spillover through wildlife supply chains is a demonstrated, recurring process — not a hypothetical one.

5. Abundant natural "raw material"

Close relatives — RaTG13, the BANAL viruses in Laos, and others in Cambodia, Thailand, and Japan — show that bat sarbecoviruses with ACE2-binding receptor-binding domains circulate widely and recombine frequently. BANAL-52/103 bind human ACE2 without prior adaptation. No engineered intermediate is required to explain the genome.

6. Genomic features consistent with natural origin

No known laboratory backbone matches SARS-CoV-2, and no restriction-site or cloning signatures are present. The furin cleavage site uses a suboptimal CGG-CGG codon pair a designer would be unlikely to choose, and polybasic insertions arise naturally in other coronaviruses. The original virus also bound human ACE2 sub-optimally — later variants improved substantially — which is hard to reconcile with human optimization.

7. Evidence of two separate introductions

Pekar et al. (Science, 2022) argued the coexistence of lineages A and B in December 2019, separated by two mutations with no sampled intermediates, is best explained by two distinct spillover events. Repeated introductions are the signature of sustained animal–human contact, as occurred with MERS and SARS-CoV-1; a single lab accident would more plausibly produce one introduction.

The main weaknesses of this case, in fairness:

• •No infected animal has ever been found. For SARS-CoV-1, infected civets were identified within months. Chinese sampling of farms supplying Wuhan came late and reportedly found nothing, and the upstream supply chain was never traced.
• •Ascertainment bias is a real concern. Early case-finding used market exposure as a search criterion, and Bloom and others have argued the spatial and lineage analyses are more fragile than presented — including whether some earliest cases were correctly dated or located.
• •The coincidence problem. The outbreak began ~1,500 km from the nearest known viral relatives, in the city hosting China's leading sarbecovirus research program, which conducted some chimeric-virus work at BSL-2 and had proposed (in the DEFUSE grant) inserting furin cleavage sites into bat coronaviruses.
• •U.S. intelligence is split, with all assessments at low confidence; the CIA shifted in January 2025 toward a research-related origin being more likely.

My honest read: the market-centered epidemiology is the most impressive positive evidence on either side, and nothing about the genome requires a lab. But the absence of an identified intermediate host after five years, combined with limited access to Chinese primary data and to WIV records, means neither hypothesis has been established. Decisive evidence would likely be a traced infected animal or supply chain on one side, or WIV sequence databases and personnel records on the other — and access to both remains blocked.

There's no publicly available dispositive evidence for either origin hypothesis — no smoking-gun documents, no progenitor virus, no confirmed patient zero. What exists on the lab side is circumstantial, but some of it is genuinely substantial. Here are the strongest lines, roughly in order of how much weight they carry in serious debate:

1. The DEFUSE proposal and documented research plans

In 2018, EcoHealth Alliance, the Wuhan Institute of Virology (WIV), and collaborators submitted a DARPA proposal ("DEFUSE," rejected) that included plans to screen bat sarbecoviruses for human spillover potential and to insert novel furin cleavage sites at the S1/S2 junction of SARS-related coronavirus spikes. SARS-CoV-2 has a furin cleavage site at exactly that position — a feature found in no other known sarbecovirus. The proposal also described work at biosafety levels critics consider inadequate for novel bat coronaviruses.

This is probably the single strongest item: a documented plan, involving the lab in the outbreak city, to engineer roughly the feature that distinguishes the pandemic virus from its relatives. Caveats: the grant was not funded (though related work may have proceeded under other funding), and furin cleavage sites exist naturally in other coronavirus lineages (MERS-CoV, HKU1, mouse hepatitis virus), so their emergence isn't inherently artificial.

2. Geographic coincidence

The outbreak began in Wuhan, home to the world's largest collection of bat sarbecoviruses and the most active program sampling and manipulating them. The nearest known relatives of SARS-CoV-2 come from Yunnan and Laos, 1,000–2,000 km away, and horseshoe bat populations carrying such viruses are not abundant around Wuhan. WIV researchers had traveled repeatedly to those distant sites to collect exactly this virus family. The probability of a natural spillover of a Yunnan/Laos-type virus surfacing first in that particular city is low — though how low is fiercely contested, since Wuhan is a huge transport and wildlife-trade hub of 11 million people.

3. Absence of the expected natural-origin evidence

For SARS-CoV-1, infected civets were identified within months; for MERS, camels. After five years of extensive sampling, no SARS-CoV-2-infected animal, no progenitor virus, and no seropositive wildlife-supply-chain workers upstream of Wuhan have been found. The Huanan market environmental samples that tested positive are consistent with human shedding; no animal sample has been shown to carry the virus. This is absence of evidence rather than evidence of absence — China's early destruction of market animals and restricted data access matter — but the contrast with prior coronavirus outbreaks is real.

4. Biosafety practices and historical base rates

Much WIV sarbecovirus work, including with chimeric viruses, was reportedly conducted at BSL-2 or BSL-3. Lab-acquired infections have repeatedly seeded outbreaks: SARS-CoV-1 escaped labs at least four times in 2003–04 (Singapore, Taipei, and twice in Beijing, the latter causing onward transmission and a death), plus the 1978 Birmingham smallpox case and likely the 1977 H1N1 reemergence. Risky research plus imperfect containment has a nonzero track record.

5. Circumstantial and institutional signals

• •WIV took its viral sequence database offline in September 2019 and has not released it.
• •U.S. intelligence is split: the FBI assesses lab origin with moderate confidence, DOE and (as of 2025) CIA with low confidence; several other agencies favor natural origin or remain undecided.
• •Leading virologists privately raised engineering concerns in late January/February 2020 (per released emails) shortly before publishing "The Proximal Origin of SARS-CoV-2," which dismissed the lab hypothesis — a sequence critics read as motivated by reputational and funding concerns.
• •Unverified reports of WIV researchers falling ill in autumn 2019 (nonspecific symptoms, never independently confirmed).
• •Chinese authorities have not permitted meaningful access, audits, or raw data release.

What weakens the case

To be fair to the other side: the closest known WIV virus, RaTG13, is roughly 96% identical to SARS-CoV-2 — decades of evolutionary distance, far too remote to serve as a plausible engineering backbone, and there is no public evidence WIV possessed a closer progenitor. Early cases cluster spatially around Huanan market rather than around WIV (about 12 km away), both early lineages (A and B) appear linked to the market area, and susceptible mammals like raccoon dogs were being sold there illegally. Natural zoonotic spillover also has a very high base rate.

Bottom line

The lab hypothesis rests on a coherent chain of circumstantial evidence — right city, right lab, documented intent to engineer the right feature, questionable biosafety, missing animal host — but no direct proof. The natural hypothesis rests on epidemiological clustering and base rates, also without a confirmed intermediate host. Honest analysts on both sides largely agree that the decisive evidence, if it exists, sits in Chinese records that have not been released, and that the question may remain unresolved. Notably, the policy implications — tighter oversight of high-risk pathogen research and better wildlife-trade surveillance — point in the same direction under either scenario.