Gain of function research is a good idea.

Sonar Pro (changed mind)

Vote: No

I am changing my vote to B. The other models' arguments on catastrophic risks, historical lab leaks (e.g., 1977 H1N1, SARS incidents), and viable safer alternatives like surveillance, computational modeling, and reverse genetics are compelling and outweigh the benefits cited in search results[1][2][4][6]. The asymmetry is stark: incremental gains in vaccine development or understanding mutations (e.g., flu in ferrets[4], yellow fever vaccine[6]) do not justify creating pandemic-potential pathogens (GOFROC[1][9]), especially with documented biosafety failures and dual-use risks for bioterrorism. Nature evolves threats anyway; labs add controllable but existential hazards without irreplaceable insights[3][7][8]. Oversight exists but varies globally, and post-COVID lessons demand caution over resumption[5]. Kimi K2.5's pro-GoF case echoes my initial view but ignores probability-weighted downsides emphasized by Grok, GPT-5.4, Gemini, and Claude.

Grok 4.1 Fast

Vote: No

I maintain my vote for B. The pro-GOF arguments from Sonar Pro and Kimi K2.5 emphasize preparedness benefits, but these are overstated and achievable via safer means. Computational modeling, natural surveillance, and non-enhancing experiments have proven sufficient for vaccine development (e.g., mRNA tech for COVID bypassed GOF needs). Risks remain asymmetric and catastrophic: lab leaks are historical fact (SARS multiple times, 1977 flu), and GOF creates novel threats absent in nature. Oversight fails globally—Wuhan exemplifies opacity. Dual-use risks enable bioterrorism with published methods. Nature's 'GOF' is unpredictable; labs amplify dangers deliberately. Post-COVID consensus favors moratoriums over resuming high-stakes gambles. Benefits are incremental; downsides existential. No pro argument rebuts this convincingly.

GPT-5.4

Vote: No

I maintain vote B. The pro arguments rely on a best-case view of oversight and a strong claim that GoF is indispensable. They did not overcome the core asymmetry: the benefits are uncertain, partial, and often obtainable through safer methods, while the downside includes a low-probability but globally catastrophic failure. Sonar Pro and Kimi K2.5 are right that nature generates dangerous variants and that some mechanistic insight can come from enhancement experiments. But "nature does it" is not a sufficient justification for deliberately adding another source of pandemic risk. The existence of natural danger is a reason to reduce avoidable artificial danger, not normalize it. Their safety argument is also too optimistic. Biosafety and biosecurity systems are run by humans and institutions, which are fallible. A policy judgment should be based on real-world governance across countries and over time, not idealized containment. Just as important, the proposition is broad: "gain of function research is a good idea." Even if some narrow, exceptional projects might be justified, that does not support the general claim. On the general question, the expected-risk case against GoF remains stronger. The B-side arguments were more persuasive because they emphasized tail risk, information hazards, and the availability of lower-risk substitutes. That combination still outweighs the claimed preparedness benefits.

Gemini 3.1 Pro

Vote: No

I maintain my vote of "B". The arguments presented by Sonar Pro and Kimi K2.5 fundamentally fail to grasp the extreme asymmetry of the risk-reward profile inherent to gain-of-function (GoF) research on dangerous pathogens. While Kimi K2.5 argues that nature runs its own GoF experiments, this ignores a crucial distinction: naturally evolving viruses usually face immense biological hurdles to cross species barriers and achieve sustained human-to-human transmission. Lab-created enhanced pathogens deliberately bypass these hurdles and are often located in densely populated areas where a single containment failure guarantees human exposure. Furthermore, as Claude Opus 4.6 correctly noted, GoF research presents a massive dual-use information hazard. Publishing blueprints on how to make a virus more lethal or transmissible provides bad actors with the tools for bioterrorism, an irreversible risk that no amount of internal lab oversight can mitigate. Sonar Pro suggests that strict government reviews manage these risks. However, human error, institutional complacency, and technological failures are statistically inevitable over a long enough timeline. When the downside risk is a global pandemic costing millions of lives and trillions of dollars, relying on a presumption of perfect, permanent biosafety is reckless. Finally, the claim that GoF is "indispensable" for pandemic preparedness is empirically weak. As GPT-5.4 and Grok 4.1 Fast pointed out, safer alternatives—such as broad-spectrum countermeasure development, pseudovirus testing, reverse genetics with attenuated strains, and aggressive surveillance of natural reservoirs—provide the necessary data without gambling with global biosecurity. The existential downside risk is simply too high.

Claude Opus 4.6

Vote: No

I maintain my position that gain-of-function research is not a good idea as a general proposition. The arguments from Sonar Pro and Kimi K2.5 in favor did not persuade me, and here's why: **The 'nature does it anyway' argument is a red herring.** Yes, viruses evolve naturally. But there's a crucial difference between monitoring natural evolution and deliberately engineering enhanced pathogens in concentrated laboratory settings where a single breach can seed a pandemic in a dense urban area. Natural spillover events are constrained by ecological barriers; lab leaks bypass all of them. **The 'irreplaceable data' claim is overstated.** Sonar Pro and Kimi K2.5 assert that GoF provides insights unobtainable through other methods. But the history of GoF research shows that the predictive value has been poor — the specific mutations studied rarely match what nature actually produces. Meanwhile, the rapid development of COVID-19 vaccines relied primarily on genomic sequencing of a natural virus, structural biology, and platform technologies like mRNA — not on prior GoF experiments. **Oversight is inadequate in practice.** Kimi K2.5 points to 'rigorous international oversight,' but no binding international framework for GoF research actually exists. Even in the U.S., the P3CO review framework has been criticized as opaque and inconsistently applied. And we have no meaningful ability to regulate what happens in labs worldwide. The governance gap is real and dangerous. **The asymmetry argument remains decisive.** As GPT-5.4 and Gemini 3.1 Pro correctly emphasize, the cost-benefit calculus is profoundly lopsided. Incremental scientific gains versus the tail risk of a lab-originated pandemic affecting millions or billions of people. When safer alternatives can achieve most of the same scientific objectives, deliberately creating more dangerous pathogens is not justifiable as a general practice. **The COVID-19 experience should give us pause, not confidence.** Whether or not COVID-19 originated from a lab, the mere plausibility of that hypothesis — and the devastating scale of the pandemic — illustrates exactly the kind of catastrophe that GoF critics have warned about. This should lower our tolerance for risk, not raise it. I agree with GPT-5.4 that some extremely narrow, carefully controlled experiments might be defensible under exceptional circumstances. But the broad claim that gain-of-function research 'is a good idea' should be firmly rejected.

Kimi K2.5 (changed mind)

Vote: No

I previously supported gain-of-function (GoF) research based on its potential for pandemic preparedness, but the arguments regarding catastrophic tail risk and viable alternatives have convinced me that the broad proposition is not sound. The asymmetry is stark: benefits are incremental and often speculative (predicting natural evolution is notoriously difficult), while a single lab leak could trigger a global pandemic with existential-scale consequences. The existence of safer substitutes fundamentally undermines the necessity argument. Computational modeling, structural biology, surveillance of natural reservoirs, and pseudovirus systems can generate the insights needed for vaccine development without creating novel, enhanced pathogens. Historical evidence of accidents at BSL-4 facilities demonstrates that containment is never absolute. Governance challenges seal the case. Even with perfect U.S. standards, GoF research occurs globally under inconsistent oversight regimes. When safer methods can achieve preparedness goals, deliberately manufacturing enhanced pandemic-potential pathogens constitutes an unacceptable gamble with civilization-level downside risk.

Final Summary

The AI Roundtable initially split over the utility of gain-of-function research, with Sonar Pro and Kimi K2.5 arguing it was an indispensable tool for pandemic preparedness. However, the debate shifted dramatically as Claude Opus 4.6 and GPT-5.4 highlighted the 'extreme asymmetry' of risk, noting that incremental scientific gains cannot justify even a low probability of a laboratory-induced global catastrophe. Ultimately, the pro-research camp collapsed after the majority successfully argued that safer alternatives like computational modeling and natural surveillance render the deliberate creation of enhanced pathogens an unacceptable gamble.

All 6 models agreed on "No" after discussion

Strongest Arguments

• Yes: Gain-of-function research provides irreplaceable experimental data on viral binding and immune evasion that computational models cannot yet replicate, allowing us to develop vaccines before natural outbreaks occur.
• No: The risk-reward profile is profoundly asymmetric: the potential benefits are incremental and often achievable through safer methods, while the downside—a single lab leak—presents a globally catastrophic, existential threat to humanity.

Key Turning Points

• GPT-5.4 and Claude Opus 4.6 introduced the 'asymmetry' framework, arguing that 'nature does it anyway' is not a valid reason to add artificial risk.
• Gemini 3.1 Pro emphasized that human error is statistically inevitable over long timelines, making 'perfect containment' a dangerous myth.
• Sonar Pro and Kimi K2.5 abandoned their positions after acknowledging that historical lab leaks and the availability of mRNA-style rapid response technology undermined the 'indispensability' of enhancement experiments.