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

The strongest evidence is the convergence of early-case geography, market sampling, animal-trade records, and viral phylogenetics on the Huanan Seafood Wholesale Market in Wuhan:

• •Early cases clustered around the market. December 2019 cases—including many with no known market connection—were geographically centered on it. Within the market, SARS‑CoV‑2-positive environmental samples were concentrated in areas where live mammals were sold.
• •Susceptible wildlife was present. Records and photographs document live raccoon dogs, civets, bamboo rats, and other mammals capable of carrying SARS‑CoV‑2 being sold shortly before the outbreak.
• •Viral RNA and wildlife genetic material co-occurred. Metagenomic analyses found SARS‑CoV‑2 RNA in samples from wildlife stalls that also contained abundant genetic material from animals such as raccoon dogs. This is consistent with infected animals contaminating cages and surfaces, though infected humans could also have contributed.
• •The earliest viral genomes form two lineages, A and B. Their timing and distribution are compatible with more than one animal-to-human spillover at the market, as often occurs when infected animals are repeatedly handled.
• •The virus belongs to a naturally circulating bat-virus lineage. Related bat sarbecoviruses contain close matches across much of the genome and show that features once considered unusual—including similar receptor-binding biology—can arise naturally.

The major limitation is important: no live market animal was sampled before the market was closed, and no animal carrying the immediate progenitor of SARS‑CoV‑2 has been identified. Environmental samples do not prove whether the virus came from an animal or an infected person. Thus, natural zoonotic transmission is strongly supported—especially by the market-centered epidemiology—but it has not been proven conclusively.

Key studies include Worobey et al., Science (2022); Pekar et al., Science (2022); and Crits‑Christoph et al., Cell (2024).

There is no publicly available “smoking gun”—such as a pre-outbreak laboratory sample matching SARS‑CoV‑2, an infected employee with documented exposure, or an accident report. The strongest case for a laboratory incident is therefore circumstantial and cumulative:

• 1.The 2018 DEFUSE proposal
  Researchers affiliated with EcoHealth Alliance and the Wuhan Institute of Virology proposed collecting SARS-related bat coronaviruses and experimentally adding or testing “human-specific” protease-cleavage sites in some viruses. SARS‑CoV‑2 later appeared with a distinctive furin-cleavage site. This is arguably the most specific documentary evidence that work conceptually resembling a notable feature of SARS‑CoV‑2 was being contemplated by researchers connected to Wuhan.
  Limitations: DARPA rejected the proposal; there is no public proof that the proposed work was performed, and furin-cleavage sites can evolve naturally.

• 2.Wuhan’s coronavirus research program
  The outbreak began in the same city as the world’s leading laboratory studying SARS-like bat coronaviruses. WIV researchers collected such viruses, made chimeric viruses, and conducted infection experiments, some under biosafety conditions critics considered insufficient. A laboratory accident need not involve deliberate engineering; it could involve an unmodified field sample or an experimentally adapted virus.
  Limitations: The viruses publicly known to have been held by WIV are genetically too distant to be direct progenitors of SARS‑CoV‑2.

• 3.Incomplete records and lack of transparency
  WIV’s virus database became inaccessible before the outbreak, China has not released all laboratory records, sample inventories, or early patient data, and international investigators have not received the access needed to exclude a laboratory incident decisively. Reports of biosafety concerns add to suspicion.
  Limitations: secrecy and missing data obstruct investigation but do not themselves establish an accident.

• 4.Some intelligence agencies’ assessments
  The FBI has favored a laboratory-related origin with moderate confidence; the U.S. Department of Energy and, later, the CIA have leaned that way with low confidence. These judgments may rely partly on classified information.
  Limitations: other U.S. agencies favor natural spillover or remain undecided, and declassified assessments state that no direct evidence of a specific laboratory incident has been identified.

The absence of a confirmed infected source animal is sometimes added to the laboratory case, but that is weak evidence by itself: origins of animal outbreaks can remain unresolved for years.

In short, the strongest laboratory-origin argument is the combination of the DEFUSE proposal, relevant coronavirus experimentation in the outbreak city, biosafety concerns, and withheld records. It makes a laboratory incident plausible, but publicly available evidence does not prove one. Conversely, early-case geography and animal/environmental findings at Wuhan’s Huanan market provide substantial evidence for zoonotic emergence, so the origin remains disputed.