New Report Maps Taxonomy of AI 'Omnicidal' Risks
24 Jul 2026
A newly published report introduces a taxonomy of potential "omnicidal events" — scenarios in which AI systems could contribute to the deaths of all or nearly all humans. The report, available at arxiv.org/abs/2507.09369, aims to support preventive measures against catastrophic risks from AI by categorizing the ways such extreme outcomes could unfold.
What the report says
According to the report, omnicidal events are defined specifically as scenarios where all or nearly all humans are killed as a result of AI. The stated purpose of the taxonomy is to give researchers and policymakers a structured way to think about and categorize these worst-case outcomes, with the ultimate goal of informing preventive measures against catastrophic AI risk.
Beyond that framing, the underlying report offers limited detail. It does not specify the exact categories or examples that make up the taxonomy, nor does it describe the methodology used to construct it. There is also no information available on the report's authors, their institutional affiliation, or whether the work has undergone peer review. No publication date is given beyond the source link itself, and the report does not lay out specific preventive measures or assess their feasibility.
Why founders should care
For founders building or deploying AI systems, this report is likely more of a signal than a directly actionable resource. It suggests that academic and research attention to extreme, low-probability-but-high-severity AI risks is growing — a trend that could plausibly shape future safety expectations or regulatory conversations, even if no such rules exist yet.
Founders may want to treat this as an early indicator rather than a mandate. It's possible that taxonomies like this one could eventually inform how regulators or enterprise customers frame AI safety requirements, though nothing in the report itself specifies concrete implementation steps. Startups that engage early with AI safety framing — even informally, by understanding how researchers categorize catastrophic risk — may be better positioned if such frameworks gain traction in policy or industry standards down the line.
What's missing
The report leaves several open questions. There's no detail on the specific risk categories or illustrative examples within the taxonomy, no clarity on who produced the work or its review status, and no discussion of proposed safeguards or how practical they might be. For founders looking to act on this, the current lack of specifics limits how directly it can inform product or policy decisions today.
Bottom line
This report represents an early, structural attempt to categorize the most extreme potential harms from AI. While it doesn't offer founders concrete guidance yet, it may be worth monitoring as a marker of where catastrophic-risk research and, potentially, future regulatory attention could be headed.