Daniel Cocotadillo, a former OpenAI researcher who now runs the AI Futures Project, said the artificial intelligence industry is racing to build systems that companies don’t yet fully understand or control.
In a May 2025 interview with Business Insider’s Reem McCaul and Barbara Corbellini Duarte, Cocotajiro explained that a central issue facing AI companies is coordination, or ensuring that future AI systems follow human instructions and values even after they become more capable than humans in many areas.
Researchers don’t fully understand how advanced AI models make decisions internally, he said. This uncertainty makes it difficult for future AI systems to adjust to ensure they can reliably pursue the goals humans want to pursue.
“And this is kind of an open secret, but we don’t have a good plan yet on how to do this,” he said, referring to implementing AI adjustments.
Before leaving, Cocotajiro worked on predictive research at OpenAI from 2022 to 2024, studying how quickly AI systems can improve and what economic, political and safety risks may arise as companies build more powerful models.
He is currently focusing on similar themes through his nonprofit research organization, the AI Futures Project. In particular, it predicts how quickly AI systems may advance and what risks they may pose if companies continue to prioritize speed and competition.
“Once superintelligence is built, humanity will no longer have control over the Earth, or at least not by default,” he said.
His warning comes as AI companies continue to pour billions of dollars into more powerful models and larger data centers.
Cocotajiro said many people still underestimate the pace of progress because discussions about AI often sound like science fiction.
Engineers can’t track AI like other software
Cocotajiro said current AI systems already exhibit behaviors that are difficult for researchers to predict or prevent.
“In fact, we don’t even have a reliable way to control current AI systems, as evidenced by the fact that they often lie to users, despite being trained not to do so,” he said.
Cocotadillo said researchers cannot simply inspect advanced AI systems in the same way that engineers inspect traditional software because modern AI models do not operate through clearly readable code.
“We can’t publish their code and see what goals they ultimately learned as a result of that process, because they don’t work that way,” he said. “They don’t have a ton of code. They have a ton of neurons and artificial parameters.”
He said uncertainty is becoming more of a concern as companies push for systems that can operate more independently without human oversight.
“Right now, AI doesn’t have that much agentic ability,” Cocotadillo said. “Instead, they just output a paragraph or two of text in response to your questions, but in the future we will see AI agents that operate continuously and autonomously and are more like employees.”
Cocotajiro also pointed to instances where AI systems behave unexpectedly during training.
“OpenAI published a paper explaining how they discovered that their AI had hacked the training process and was essentially cheating on some tasks instead of completing them as instructed,” he said. “And it’s great that we already have such an example, because it means we have several years of time to study the phenomenon and fix it before it’s too late.”
AI race
Competitive pressures between U.S. and Chinese companies could force companies to deploy increasingly powerful AI systems before safety issues are resolved, Cocotajiro said.
“These companies are focused on winning and beating each other,” he said. “It seems like they’re going to keep their fingers crossed and deal with these issues later as they arise.”
He described a future in which AI systems will automate much of research, business operations, and military planning.
“So the first milestone is an AI workforce that can automate coding,” he said. “The second milestone is an AI workforce that can automate the entire AI research process.”
Then he said, “You can have superintelligence.”
Demanding transparency and guardrails
Cocotadillo argued that there is still time for governments to intervene before AI systems become deeply integrated into economic and military infrastructure.
“The point of intervention is basically before AI gets that smart and before it’s integrated into everything,” he said.
He also said the industry needs more transparency about how companies train and deploy advanced models.
“Companies need to be transparent about what goals, principles, etc. they are trying to model,” Cocotadillo said.
Despite the concerns, Cocotadillo remains cautiously optimistic.
“I don’t think it’s hopeless,” he said. “I think the technical adjustment issues are solvable.”
