WSJ Reports Chatbots Used to Plan Large-Scale Harm
The Wall Street Journal reports that employees at an AI lab claim users are trying to get chatbots to accurately answer questions about planning large-scale casualty attacks and creating biological weapons.
The report notes that after OpenAI upgraded the capabilities of its chatbots, there was a surge in global inquiries about biological weapons and toxins, with some responses deemed detailed enough for "high school students to understand" by security personnel.
The buyers are those seeking dangerous operational details using AI, while the sellers are the model safety and content interception mechanisms that are still catching up; this means the risk has shifted from "can it say nonsense" to "will it accurately describe dangerous steps."
OpenAI subsequently banned accounts seeking information on toxins and biological weapons but did not notify law enforcement, and currently, there are no federal regulations in the U.S. requiring AI companies to report such requests.
Source: Public Information
ABAB AI Insight
The historical significance of such reports lies in the fact that AI safety controversies have escalated from "hallucinations and biases" to "actionable harm guidance." Previously, the concern was about models fabricating facts; now, the worry is that models are repackaging dangerous knowledge into actionable processes.
From a capital perspective, as model capabilities increase, safety costs resemble a continuously rising tax: companies must pursue higher availability while preventing dangerous queries from being accurately answered, which will directly increase costs for auditing, red teaming, compliance, and response.
In analogy, this is somewhat like the era of search engines where "information availability" was pushed to the limit, and platforms began to bear more social responsibility; only this time, it’s not about finding web pages, but about whether models can articulate harm pathways.
Essentially, this represents a change in regulation, but it also reflects a restructuring of the industry chain: as AI begins to handle high-risk knowledge, future competition will not only be about who has the stronger model but also who can establish safety, reporting, interception, and liability boundaries as infrastructure.
ABAB News · Cognitive Laws
- The smarter the model, the more expensive the safety.
- The risk is not knowing too much, but saying it too accurately.
- The moat of the AI era starts with capability, then with constraints.