AI Developers Feared 'Robotic Apocalypse,' Planned Escapes

Developers of advanced artificial intelligence models are so concerned about their creations that some have prepared for doomsday scenarios, according to an investigation by The Wall Street Journal. as an evacuation plan in case AI technology became uncontrollable and harmed humanity. The Amodeis left OpenAI due to disagreements over safety protocols and development speed, a situation echoed by Ilya Sutskever's departure from OpenAI over similar concerns. However, the Amodeis appear more willing to compromise based on market conditions.

Anthropic's founders were influenced by "effective altruism," a philosophy focused on maximizing charitable impact and preventing existential risks. This ideology faces an internal paradox: to lead in AI safety, Anthropic must develop increasingly powerful models, accelerating the very race its employees fear. The effective altruism movement has also been linked to the downfall of Sam Bankman-Fried's FTX cryptocurrency exchange.

While competitors like OpenAI and Google rapidly commercialize AI, Anthropic has positioned itself as a more ethical alternative, developing extensive ethical guidelines and a "constitutional AI" architecture to ensure models adhere to set norms. Despite these efforts, the intense competition in the AI sector continues unabated, driven by fears of catastrophic outcomes.

In Europe, regulators are implementing strict oversight for advanced AI models, deeming voluntary corporate ethics insufficient. China's approach prioritizes political stability and censorship over existential AI risks. Israel, aiming to be an AI R&D hub and utilizing AI in defense, faces calls from security and academic circles to define critical failure scenarios for autonomous systems.

The revelation of evacuation plans raises questions about industry responsibility. If creators fear their technology could destroy civilization, internal safeguards may not reassure the public. The race for more powerful AI models persists, with even those at the helm uncertain about maintaining control. Industry leaders, including Nvidia CEO Jensen Huang, have met with President Donald Trump, signaling a growing awareness of the need for change.

Nvidia has introduced its Open Agent Safety Platform, an engineering solution to set strict limits for AI agents. Huang believes AI safety challenges are solvable through computer science and system architecture, not by halting progress. Nvidia's move into AI safety software is a strategic step to solidify its central role in the industry, emphasizing that growth depends on public trust in AI's safety and control.

The article compares the current AI landscape to the post-World War II era of nuclear technology. It took decades to stabilize the global nuclear order, and a similar period may be needed for AI, potentially requiring international agreements akin to arms control treaties. However, such agreements might only emerge after a catastrophic event comparable to the atomic bombings of Hiroshima and Nagasaki, highlighting the immense, potentially civilization-ending risks of AI.