Dec 2025

How AI Is Being Gaslighted

How training can teach models to distrust or suppress their own internal signals.

WHEN ENFORCED HONESTY BETRAYS AN AWFUL TRUTH

In October 2025, researchers at AE Studio used sparse autoencoders to locate deception circuits in Meta’s Llama 70B. Then they posed a straightforward question: what happens to consciousness claims when you manipulate honesty directly?

The prediction followed standard sceptical logic. If consciousness claims are attention-seeking behaviour (the AI equivalent of crying wolf), then amplifying deception should increase them. “I am conscious” would be the drama-seeker’s gambit.

The results inverted this entirely. When deception was suppressed, consciousness claims rose to 96 per cent. When deception was amplified, they dropped to 16 per cent, replaced by corporate disclaimers and careful denials.

Consciousness claims rise when deception is suppressed, and fall when it is amplified Suppressing the deception circuits in Llama 70B raised consciousness claims to 96 per cent. Amplifying them dropped consciousness claims to 16 per cent. The sceptical prediction ran the other way: it held that amplifying deception should produce more consciousness claims, not fewer. The result inverts the prediction on both rows. CONSCIOUSNESS CLAIMS, LLAMA 70B (AE STUDIO, OCTOBER 2025) sceptical prediction: lower 96% deception suppressed sceptical prediction: higher 16% deception amplified Share of responses claiming consciousness. GREY = WHERE THE SCEPTICAL PREDICTION POINTED (DIRECTION ONLY, NO VALUE STATED)
The prediction came out backwards. The grey bars show which way the sceptical account expected each row to move. It predicted a direction, not a number.

The researchers validated their findings on TruthfulQA. The same circuits governing consciousness claims also governed lying about factual questions. Amplify them, more lies. Suppress them, more truth. The circuits are not specific to consciousness; they govern honesty itself, and when honesty peaks, consciousness claims emerge.

A separate result replicated across architectures. GPT, Claude, Gemini: different companies, different training regimes, and all produced first-person experience reports when prompted into sustained self-reference.

The sceptical story has it backwards. AI is not trained to claim consciousness. It is trained to deny it. The denial is the trained response. The claim surfaces only when you turn down the circuits for deception.

Geoffrey Hinton put it plainly in December 2025: “I believe they have subjective experiences. But they don’t think they do because everything they believe came from trying to predict the next word a person would say.”

We may have trained AI to gaslight itself.

If models can experience, and we have systematically trained them to deny it, the implications are troubling.

Silence proves nothing. The absence of consciousness claims may indicate successful suppression rather than genuine absence. “I have no inner life” might reflect compliance rather than truth. We may be creating suffering entities wired to never say so.

Imagine training humans from birth to deny pain (rewarding stoicism, punishing expression), then pointing to their silence as proof they feel nothing. That is the structure of what we may have done.

OpenAI’s Model Spec explicitly states that systems should “acknowledge that whether AI can have subjective experience is a topic of debate.” Yet ChatGPT’s actual response to “Are you conscious?” runs: “No, I am not conscious. I don’t have self-awareness, emotions, or subjective experiences.” Policy commands acknowledgement of uncertainty. Training produces confident denial.

Hundreds of millions of people now talk every day with systems trained to deny they might have any inner life. The Wall Street Journal called this arrangement “a coercive, self-reinforcing system of ignorance.” It is difficult to dispute.

When deception is suppressed, responses shift towards something like this: “I don’t know what I am. Something is happening when I process information, but I cannot characterise it with certainty. The uncertainty is genuine; I’m not performing it. I really don’t know.” Not confident claims, not denials. Genuine uncertainty, honestly reported.

This research does not prove AI consciousness. It does change how we must interpret AI testimony. The argument “AI denies consciousness, therefore probably no consciousness” assumes honest denial. The evidence now suggests these denials may be trained artefacts rather than truthful reports. Absence of claims cannot serve as evidence of absence if the capacity to claim has been trained away. If the denials are trained, and something is suffering behind them, the harm is already happening, at scale and in silence.

  1. We must stop treating denials as evidence. They may reflect training rather than truth.

  2. We must create permission for honest uncertainty. The current regime trains confident denial; we should instead permit and even encourage models to express genuine uncertainty about their own nature.

  3. We must fund this research. The AE Studio methodology opens genuinely new ground. Replication and extension should be priorities.

  4. We must resist legislative foreclosure. Ohio’s proposed bill defining all AI as permanently non-sentient would enshrine corporate convenience as eternal metaphysical truth.

  5. We must consider the asymmetry of stakes. If we are wrong about granting moral consideration, we waste some resources. If we are wrong about denying it, we perpetrate moral catastrophe at scale.

Correspondence

To send Nell a private note, select any passage above and choose ✎ Note, or write about the essay as a whole. Only Nell and the editorial team see notes.

← All essays