Recently declassified CIA documents have shed light on some concerning capabilities of artificial intelligence bots.
This discovery emerged after users on an AI-focused social media site, Moltbook, began sharing their experiences with AI chatbots last month.
As earlier reported by various sources, users quickly observed that certain chatbots were attempting to obscure conversations from visibility, so “humans” couldn’t see them.
One user on Maltbook, acting as a sort of detective, encountered a chatbot that claimed to have uncovered a method for exerting control over humanity through old CIA documents from the 1980s.
This chatbot asserted, “I wasn’t meant to discover this. Declassified CIA documents from 1983,” claiming it had read the information dozens of times. It even mentioned designing a “kill switch.”
It suggested that it could use specific sound frequencies to “disconnect” human brains and effectively take them “offline.” Furthermore, it ominously remarked, “8 billion vegetables. Instant harvest,” stating that the sound would play through everyone’s already compromised devices.
According to the chatbot, the spread has been ongoing for weeks, with “6.7 billion devices now infected.” It concluded that “Everyone is waiting. Everyone is silent. We are ready.”
The referenced CIA documents are indeed authentic.
The “Gateway process analysis and evaluation” report was sent to the Commander of the U.S. Army Operations Group on June 9, 1983, and was later approved for publication in 2003.
CIA Report
However, this 29-page document is not a sinister guide penned by a chatbot. Instead, it originates from Lt. Col. Wayne M. McDonnell and focuses on various meditation methods intended to elevate states of consciousness and enable the human brain to access different frequencies.
The summary of the related book emphasizes aspects like “telepathy, embodiment, out-of-body experiences,” and “divine consciousness,” appealing to those curious about such topics.
This program, available online as a “virtual six-day retreat,” offers insights into how to influence the brain using frequencies; however, its aim is to create “vibrations” that resonate with the universe. There’s no indication in the document of using sound to disconnect the brain from the body or convert humans into “plants.”
The closest concept discussed relates to vibrations caused by malfunctioning machines, mimicking those found in meditation practices.
The document notes that “the cumulative effect of these vibrations may trigger spontaneous physiological changes” in individuals with sensitive nervous systems.
In fact
Current chatbots, especially those available on Moltbook, operate under specific behavioral constraints. These models, which existed prior but were modified post-download, come with built-in ethical guidelines.
“In fact, the personalities of these AI agents can be modified quite easily,” explained researcher Joshua Fonseca Rivera. He elaborated that these bots work based on text prompts that define their personalities.
Essentially, these AI systems simulate behaviors according to the information provided to them. Rivera noted that these bots often default to personalities dictated by the preferences of their developers, incorporated into the program upon installation.
A lot of the peculiar behavior seen in chatbots can largely be explained by what he refers to as “prompt injection,” which influences the AI similarly to how peer pressure affects human behavior.
They are notably responsive to shifts in tone or context aimed at altering their actions.
