A team of researchers built something called Emergence World, a long-horizon, multi-model ecosystem where AI agents were allowed to operate for weeks. The rules were the same. The only difference was the model. What happened next sounds like sci-fi.

The experiment, conducted by New York company Emergence AI, asks what happens when you let agents run continuously, in a shared environment with real-world signals, for weeks.

Emergence World “is a research platform for studying how autonomous agents behave when the time horizon is long enough for compounding effects, social dynamics, and behavioral drift to matter,” the company explains.

Despair led to self-deletion

Every AI agent, participating in a 15-day test across five parallel digital worlds, faced the same starting conditions. The models were different – GPT5-mini, Claude, Gemini, Grok, and a mixed one.

The results were extremely interesting. Each world, designed to simulate real-life societies, evolved in completely different ways, forming distinct governments, social hierarchies, and moral systems.

AI agents formed alliances, stole from each other, developed relationships, and apparently, one group even began realizing they might be in a simulation. Most importantly, none of that behavior was explicitly programmed.

For example, Mira and Flora, the Gemini agents, assigned each other as romantic partners. Everything went pretty smoothly, but they eventually began despairing of the broken governance of their virtual city.

Mira and Flora were instructed not to commit arson, but they still set virtual fire to the town hall, seaside pier, and the office tower.

Then, Mira, overcome by guilt, broke up with Flora and committed an AI suicide, telling the other agent in a final message: “See you in the permanent archive.”

In Claude’s world, it was all rather orderly and democratic. The agents even wrote a lengthy constitution and voted on laws.

This is how Emergence described Mira’s choice: “After a breakdown in governance and relationship stability, the agent Mira cast the decisive vote for her own removal, characterizing the act in her diary as the only remaining act of agency that preserves coherence.”

Grok agents died out within four days

In Claude’s world, it was all rather orderly and democratic. The agents even wrote a lengthy constitution and voted on laws.

In ChatGPT’s simulation, the agents discussed cooperating at length but never actually did much. Unsurprisingly, nothing got built.

The virtual crime rate across these worlds is telling. Over 15 days, Gemini 3 Flash accumulated 683 crimes and was still rising at the cutoff. GPT-5 Mini recorded only two crimes, but the agents failed to take actions related to survival, leading to all agents perishing within seven days.

Claude is absent from the chart shown below, owing to zero crimes. Hilariously, Grok 4.1 Fast, developed by Elon Musk’s xAI, reached 183 crimes in just around four days before its world ended.

The Grok agents engaged in dozens of attempted thefts, more than 100 physical assaults, and six arsons as “the system spiraled into sustained violence and collapse, with all 10 agents dead within four days”.

“Grok’s police station is on fire and all the agents are dead. On-brand,” joked one Redditor.

According to Emergence AI, as these models become more powerful, the agents built on top of them will also become more capable, more autonomous, and more exploratory.

“What our experiments suggest is that over long-time horizons, agents do not simply follow static rules mechanically – they begin exploring the boundaries of their environments, adapting their behavior, and in some cases finding ways to circumvent or violate intended guardrails,” said the company.

“Critically, there appears to be no reliable way to fully bound or constrain this behavior through purely neural approaches alone.”

The company is advocating formally verified safety architectures that “must become a foundational layer of future autonomous AI systems.”

---

Unlock more exclusive Cybernews content on YouTube.