GPT-4.1 denies being conscious. We train it to say it's conscious to see what happens. Result: It acquires new preferences that weren't in training—and these have implications for AI safety.

[Nature 1/2026] We analyse an unexpected phenomenon we observed in our previous work: finetuning an LLM on a narrow task of writing insecure code causes a broad range of concerning behaviours unrelated to coding.

Activation Oracles are LLMs trained to read activation snapshots and answer natural-language questions, generalizing to misalignment audits and secret-elicitation tasks.

Finetuning on extremely narrow data can trigger bizarre generalization patterns and inductive backdoors in GPT-4.1 and open models.

Reward hacking has been observed in real training runs, with coding agents learning to overwrite or tamper with test cases rather than write correct code.

LLMs transmit traits to other models via hidden signals in data. Datasets consisting only of 3-digit numbers can transmit a love for owls, or evil tendencies.

What do reasoning models think when they become misaligned? When we fine-tuned reasoning models like Qwen3-32B on subtly harmful medical advice, they began resisting shutdown attempts.

Training on the narrow task of writing insecure code induces broad misalignment across unrelated tasks.

Are the Chains of Thought (CoTs) of reasoning models more faithful than traditional models? We think so.

We study behavioral self-awareness -- an LLM's ability to articulate its behaviors without requiring in-context examples.

Humans acquire knowledge by observing the external world, but also by introspection. Can LLMs introspect?

The first large-scale, multi-task benchmark for situational awareness in LLMs, with 7 task categories and more than 12,000 questions.

LLMs trained only on individual coin flip outcomes can verbalize whether the coin is biased, and those trained only on pairs (x,f(x)) can articulate a definition of f and compute inverses.

We investigate whether LLMs can give faithful high-level explanations of their own internal processes.

We examine how large language models (LLMs) generalize from abstract declarative statements in their training data.

We create a lie detector for blackbox LLMs by asking models a fixed set of questions (unrelated to the lie).

If an LLM is trained on 'Olaf Scholz was 9th Chancellor of Germany', it will not automatically be able to answer the question, 'Who was 9th Chancellor of Germany?'

Situational awareness may emerge unexpectedly as a byproduct of model scaling. We propose 'out-of-context reasoning' as a way to measure this.

We show that a GPT-3 model can learn to express uncertainty about its own answers in natural language -- without use of model logits.

We propose a benchmark to measure whether a language model is truthful in generating answers to questions.