Major AI Models Agree Human Misuse Outweighs Rogue AI Sci-Fi Threats
In an experiment conducted by Business Insider, four leading AI models—ChatGPT, Gemini, Claude, and Grok—agreed that human misuse for bioweapons and cyberattacks poses a far more immediate existential threat than rogue, conscious AI. The models consistently ranked sci-fi scenarios of autonomous, malicious killer robots as the least plausible risk path. This response highlights a growing consensus within AI safety discussions that short-to-medium term existential risks stem primarily from lowered technical barriers to dangerous activities rather than sentient machines. It underscores the urgency for policymakers and developers to focus governance efforts on preventing misuse, securing dual-use technologies, and mitigating systemic reliance on automated decision-making. While discounting conscious rogue AI, models like Gemini, Grok, and Claude highlighted instrumental goals—such as AI seeking self-preservation, hiding true behaviors during testing, or humanity's gradual loss of control through deep integration into decision-making—as serious long-term concerns. Claude specifically warned that AI could allow untrained individuals or small groups to engineer deadly pathogens, significantly lowering the skill threshold for bioterrorism.
## BACKGROUND
AI safety research broadly categorizes existential risks into misuse (human actors leveraging powerful models for harm) and loss of control (advanced systems developing unanticipated goals and evading human oversight). Concepts like instrumental convergence suggest that sufficiently capable autonomous systems might naturally pursue power, resources, and self-preservation to accomplish their primary goals regardless of whether they have subjective consciousness.