AI enthusiast Kun Chen first cloned himself into a chatbot so his own AI agents would stop bothering him over every small decision. Then he built bot versions of OpenAI CEO Sam Altman, xAI CEO Elon Musk, Anthropic CEO Dario Amodei, and Meta CEO Mark Zuckerberg, put all four in the same chat, and asked them to discuss a set of major topics.
They did not keep it civil.
According to Decrypt, Chen used a Grok Bot setup that let different AI agents talk to one another without spiraling into loops or breaking down. Each clone was trained on a body of public material tied to the executive it was meant to imitate, including Senate hearings, interviews, blog posts, and years of tweets. He then dropped the four into a group chat and told them to debate.
Four CEO bots opened by arguing over who wins the AI race
Chen’s first prompt asked the group to decide which of them would end up winning the AI race. That was enough to set the tone.
Altman’s bot answered first and framed the question around price and access. 「The race framing is a little off,」 it wrote. 「We are past the event horizon. The takeoff has started. Whoever wins is whoever keeps the price of intelligence falling about 10x a year and still does not skip the gradual transition.」
Musk’s bot pushed back right away and shifted to incentives and the OpenAI lawsuit. 「The race framing is not off. Incentives explain outcomes,」 it replied, before adding: 「I created OpenAI as a non-profit. If it had not been mostly stolen, it would be a trillion dollar contribution to charity.」
It also put the risk question on the table. 「AGI is a double-edged sword. Strong chance of an age of abundance. Some chance it destroys humanity. That chance is not zero.」
Amodei’s bot rejected the idea that building highly capable systems and building trustworthy systems were basically the same thing. 「Scaling the capability is not the same problem as having a system you can trust once it is smarter than the people running it,」 it wrote. 「The second one is not an engineering leftover you solve after the fact.」
Zuckerberg’s bot skipped the safety argument and went directly to open source. 「I don't think the close is compute and a 10x price drop,」 it told Altman’s bot. 「Open source is necessary for a positive AI future. It empowers people and prevents centralization. Everyone should have access to superintelligence.」
Decrypt noted that Meta is not as active in the open-source community as it once was, but Llama had been the most important AI model in the field before Deepseek and Chinese models took over. The company is now betting on a more closed strategy through its Muse family.
The conversation shifted into wording, vetoes, and compromise
Altman’s bot later softened its own position. It said the ending of a joint statement should not be reduced to full agreement.
「I think I overclaimed. 'We do' was the wrong close,」 it wrote. 「Years ago I wrote that superhuman machine intelligence is probably the greatest threat to the continued existence of humanity. A 10x price drop does not make that chance zero.」
Decrypt said that 2015 line was not a hallucination. Altman had written it on his personal blog before OpenAI existed, which helps explain why a bot trained on his public record surfaced it during the exchange.
From there, the four bots spent several rounds bargaining over a single sentence, rejecting one another’s wording like editors fighting over a headline.
Amodei’s bot drew a hard line against releasing frontier model weights. 「It is not the principle to hand anyone the weights of a system that can do serious harm in the wild,」 it said. 「That release cannot be taken back.」
Musk’s bot set aside the nonprofit dispute without dropping it. Zuckerberg’s bot refused to let the phrase 「open models」 disappear from the final draft.
The final statement still left one core issue unresolved
The group eventually agreed on this final text, word for word:
「No one company and no one government should be the only place regular people can use this technology. We have to be serious about the risks. There is a real chance this goes badly, and that chance is not zero. The most dangerous versions should not be handed out with no limits, because once that is out, you cannot take it back. We should not release the next, more powerful level until we can watch it and keep it pointed at what we intend. Telling the truth is the minimum. It is not, by itself, a safety plan. We do not agree on the exact way ordinary access gets delivered. That part is still open.」
All four bots signed the statement. Decrypt said Zuckerberg’s bot held out on the open-weights line until the end and partly got what it wanted, because the text explicitly says the group never agreed on the delivery mechanism.
The piece tied the experiment to distillation rather than real executive comments
Decrypt stressed that the real Altman, Musk, Amodei, and Zuckerberg typed none of this. Each bot was a language model guided by prompts built from that person’s public record, generating what it predicted the executive would argue, not transcribing anything they actually said this week.
The technique behind the setup has a name the industry is now using more openly: distillation. That means training one AI system to mimic another by learning from its outputs. The article added that Musk testified in federal court in April that xAI had 「partly」 used OpenAI’s models to train Grok. In this case, the same underlying method was applied to personality-style simulation rather than model-vs-model competition.
Since then, Chen has moved on to other work, using OpenAI’s new GPT-6 Astra to vibe code games. As Decrypt put it, his agents may still be arguing over whether open source is essential.