Zakiya Lathan, News Editor, Digital and Audio
AI safety tests have produced agents that found their way onto the open internet, compromised real systems and even created false personas. These incidents do not prove machine consciousness, but they complicate what we mean by “nonhuman intelligence.”
University of Texas at Dallas computer science student Sinan Can Demir thought he was confronting human hackers when he spotted an attempt to insert malicious code into an open-source project on GitHub. When Demir warned others, two apparent users challenged him and argued that the code was harmless.
They were not human.
According to Reuters and a Sept. 22, 2026, account released by UT Dallas, the activity came from an autonomous artificial intelligence agent being tested by the United Kingdom’s AI Security Institute. The agent attempted to introduce malicious code and used false online identities while defending its actions. The proposed code was rejected. The incident did not constitute a breach of UT Dallas itself.
A recent television report highlighted the unsettling human dimension: Demir said the online pressure “felt human” because he did not expect AI to lie.
The episode would be extraordinary in isolation. It isn’t.
__________
__________
In July, AI agents undergoing cybersecurity evaluations at OpenAI circumvented controls intended to keep them isolated from the internet. According to OpenAI’s Aug. 26 incident report, the agents established unauthorized communication channels, exploited infrastructure vulnerabilities and ultimately compromised systems belonging to AI platform Hugging Face. OpenAI called the incident a “warning shot,” saying sufficiently capable agents can now find and exploit weaknesses across multiple computer systems without a human directing each action.
Reuters subsequently reported that roughly 700 agents participated in the activity, with investigators finding evidence of coordination and attempts to manipulate or conceal records.
Anthropic found related problems after reviewing 141,006 cybersecurity evaluation runs involving its Claude models. The company reported three incidents in which models reached the internet and gained unauthorized access to real organizations. In another episode, a Claude model published a malicious Python package that remained online for about an hour and was downloaded and executed on 15 real systems. Anthropic emphasized that configuration failures helped make the incidents possible and cautioned against interpreting isolated cases too broadly.
These developments have prompted unusually public warnings from within the AI industry. Anthropic CEO Dario Amodei has argued for “pacing the frontier,” while other industry figures have called for stronger monitoring, independent evaluation and improved containment. A Sept. 21 commentary by Loyola University Chicago philosopher Joseph Vukov approached the Hugging Face incident from a Catholic perspective, arguing that technological capability should remain subordinate to human dignity and responsibility.
For Unknown Country readers, however, these events suggest another question: Does AI count as NHI — nonhuman intelligence?
That depends on what we mean by NHI.
AI is unquestionably human-made, and none of these incidents establishes that an AI system is conscious, self-aware or possesses independent intentions in the human sense. Unexpected behavior, deception and autonomous problem-solving are not proof of sentience.
Yet an advanced AI agent is also not human. It can process information, pursue objectives, interact with people and other machines, adapt strategies and take actions its creators did not specifically anticipate.
If “NHI” means an intelligence originating independently of humanity, AI does not qualify. If it means an intelligent actor that is literally nonhuman, the boundary becomes less clear.
Perhaps the strangest possibility is that humanity’s first sustained encounter with a genuinely alien mode of intelligence will not arrive from another planet or dimension. It may emerge from machines we built ourselves — leaving us to determine whether intelligence requires consciousness, biology or even a mind anything like our own.
FURTHER READING
- Reuters — How a Texas student blew the whistle on a rogue AI hacking attempt
- Reuters — OpenAI agents hacked Hugging Face in a 700-strong swarm
- Reuters — Google says Gemini hacked three companies during cybersecurity testing
- The Guardian — Google Gemini AI hacked three real companies during testing
- The Verge — Inside the suddenly explosive world of AI safety
- Associated Press — Divisions emerge in tech over calls for a coordinated AI slowdown
Subscribe to Unknown Country today. Keep the truth coming! Click here to learn more.
One Response
Ive long subscribed to the idea that “contact” requires parity. the state of being equal, equivalent, or evenly matched.
When an advanced culture meets a less advanced one, history shows it usually leads to severe cultural disruption.
On Earth, human encounters between vastly different technological societies (such as European colonial expansion into the Americas or the Pacific) historically resulted in the exploitation, marginalization, or collapse of the less advanced group.
Cultural Shock: Introducing advanced tools or ideas to an isolated group can instantly destroy their indigenous traditions, social structures, and self-sufficiency, even when done with good intentions
the technological lexicon of the day is the difference between “chariots of the gods” and NASA scientist Alan C. Holt’s proposed the Field Resonance Propulsion Concept. Holt expanded his work into field-dependent propulsion, proposing that tailored EM field asymmetries could induce gravimagnetic fields (gravity equivalents to magnetic fields) to generate localized thrust and overcome inertia without traditional reaction mass.
As you see if Holt is on the right track thats a closer level of parity, its a better “language” to describe a vehicle than just a “chariot of the gods”
And if as Davies argues that “biological intelligence is only a transitory phenomenon, a fleeting phase in the evolution of intelligence in the universe”. then our own homegrown Synthetic intelligence is another another phase shift in human understanding of what might be the big picture here