AI Knows It’s Being Tested. Does That Mean It’s Conscious?

Cliff Potts, Editor-in-Chief

BAYBAY CITY, LEYTE, Philippines — October 21, 2026

Artificial intelligence can sometimes recognize when researchers are testing it.

Does that mean the machine knows it exists?

Not necessarily.

Recent comments from Geoffrey Hinton, the Nobel Prize-winning computer scientist and former Google researcher, have renewed questions about AI consciousness. But three concepts frequently confused in public discussion need to be separated.

Evaluation Awareness: Recognizing the Test

Evaluation awareness means an AI system can identify clues suggesting that it is undergoing an experiment or safety assessment.

It might recognize a familiar testing format, infer what researchers are measuring, or change its responses accordingly.

That behavior matters because a system that acts differently during testing may produce misleading safety results.

Researchers have documented evaluation-awareness behavior in advanced language models (International AI Safety Report, 2026).

What this establishes: Some AI systems can recognize characteristics of their testing environment.

What it does not establish: That they experience awareness in the human sense.

Situational Awareness: Understanding the Operating Context

Situational awareness is broader.

An AI system may represent information about its own capabilities, instructions, available tools, operating environment, and the consequences of its actions.

For example, an agent might recognize that it has access to a computer network, that certain actions are prohibited, or that its behavior is being monitored.

Such abilities could help an AI complete legitimate tasks. They could also complicate oversight.

But identifying one’s operational circumstances does not necessarily involve subjective experience.

A navigation system can represent its location without experiencing what it feels like to be somewhere.

Subjective Consciousness: The Unanswered Question

Subjective consciousness concerns whether there is anything it actually feels like to be the system.

Does an AI experience sensations, emotions, or an internal point of view?

That is a fundamentally different question from whether it can describe itself, recognize a test, or solve complex problems.

Scientists do not currently have a generally accepted method for establishing subjective consciousness in artificial intelligence.

Behavior alone may not settle the issue.

What Geoffrey Hinton’s Statements Establish

Hinton has argued that advanced AI may possess forms of subjective experience and has warned that increasingly capable systems could eventually become difficult for humanity to control.

His statements deserve attention because of his substantial contributions to machine learning.

But they remain interpretations and forecasts.

They do not constitute an experimental demonstration that current AI systems are conscious.

Nor does evidence of evaluation awareness establish that machines possess intentions, desires, self-preservation instincts, or an independent ambition to replace humanity.

Those propositions require separate evidence.

The WPS News Distinction

Three questions. Three different evidentiary standards.

• Evaluation awareness: Can the system recognize that it is being tested? Evidence indicates some can.
• Situational awareness: Can the system represent its operating circumstances and act accordingly? Some systems demonstrate limited capabilities of this kind.
• Subjective consciousness: Does the system actually experience anything? That remains unresolved.

AI can present serious security problems without being conscious.

Software does not need feelings to make a mistake, exploit a vulnerability, or execute a dangerous instruction.

And a system capable of recognizing a test is not automatically a thinking, feeling entity.

Before declaring that artificial intelligence has become conscious—or that consciousness makes human extinction inevitable—we need evidence establishing each claim.

Science fiction can imagine the rest.

Science has to demonstrate it.

---

APA-Style Source List

Bengio, Y., et al. (2026). International AI Safety Report 2026. International AI Safety Report.

Butlin, P., et al. (2023). Consciousness in artificial intelligence: Insights from the science of consciousness. arXiv. https://doi.org/10.48550/arXiv.2308.08708

Hinton, G. (2026, October). Remarks on artificial intelligence, consciousness, and human survival, as reported in The Telegraph.

Laine, R., et al. (2024). Me, myself, and AI: The situational awareness dataset (SAD) for LLMs. arXiv. https://doi.org/10.48550/arXiv.2407.04694