AI Researchers Warn Human Misuse Outweighs Rogue Threats
A survey of AI researchers finds that human misuse and centralized corporate control pose far greater near-term risks than self-directed software.
- By Jesse Jacobs
- Oct 30, 2026
Human actors misusing software tools present a far more immediate danger than hypothetical scenarios involving rogue artificial intelligence systems, according to a recent survey released by Cybernews.
The findings poll 103 computer scientists and machine-learning researchers across Europe, Asia and North America. Approximately 38% of respondents identified intentional human misuse—including state-sponsored cyberattacks, biological weapons development and automated disinformation campaigns—as the single greatest potential source of harm over the next decade.
Another 28% of specialists cited systemic societal threats, such as centralized corporate power, institutional destabilization and severe economic disruption, as primary hazards. By contrast, roughly 20% viewed system misalignment, where an autonomous model pursues unintended goals outside user control, as the chief threat.
Despite public messaging from commercial technology vendors suggesting human-level synthetic capability is imminent, academic and independent researchers expressed measurable skepticism regarding current technical roadmaps.
More than 52% of respondents indicated that achieving artificial general intelligence—software capable of matching skilled humans across generalized intellectual tasks—requires fundamental scientific breakthroughs rather than merely scaling existing algorithmic frameworks. When assessing artificial superintelligence that substantially exceeds human expertise, roughly 70% reported that current developmental approaches remain insufficient.
The findings also reflect shifting regulatory perspectives within technical research institutions. Faced with persistent security vulnerabilities, nearly 40% of surveyed specialists advocated for slowing the pace of frontier model deployment, while approximately 16% favored a temporary development pause.
Nearly 64% of respondents nonetheless noted that self-improving algorithms could plausibly present severe systemic risks to humanity within five years if safety standards fail to keep pace with deployment.