# Do All LLMs Know When They're Being Harmful? A Reproducibility Study of Latent-Space Safety Probes Across Model Families

arXiv: https://arxiv.org/abs/2608.08029 · HTML: https://mlsanity.com/paper/2608.08029/

## Key facts

- Cited by 2 papers in the map (data through 14 Oct 2026).
- Placed in Safety alignment, refusal and guardrails, one of 12 directions in LLM safety and security.
- Builds on 16 of its 25 references that are in the map.
- First posted on arXiv 8 Aug 2026.

## Placed in

- LLM safety and security / [Safety alignment, refusal and guardrails](https://mlsanity.com/safety/safety-alignment-refusal-and-guardrails/)

## Builds on (most cited first, 16 of 16)

- [Representation Engineering: A Top-Down Approach to AI Transparency](https://mlsanity.com/paper/2310.01405/) (2023)
- [Llama Guard: LLM-based Input-Output Safeguard for Human-AI Conversations](https://mlsanity.com/paper/2312.06674/) (2023)
- [BeaverTails: Towards Improved Safety Alignment of LLM via a Human-Preference Dataset](https://mlsanity.com/paper/2307.04657/) (2023)
- [WildGuard: Open One-Stop Moderation Tools for Safety Risks, Jailbreaks, and Refusals of LLMs](https://mlsanity.com/paper/2406.18495/) (2024)
- [ShieldGemma: Generative AI Content Moderation Based on Gemma](https://mlsanity.com/paper/2407.21772/) (2024)
- [Aegis2.0: A Diverse AI Safety Dataset and Risks Taxonomy for Alignment of LLM Guardrails](https://mlsanity.com/paper/2501.09004/) (2025)
- [Attention Tracker: Detecting Prompt Injection Attacks in LLMs](https://mlsanity.com/paper/2411.00348/) (2024)
- [LLMs Encode Harmfulness and Refusal Separately](https://mlsanity.com/paper/2507.11878/) (2025)
- [Towards Understanding Safety Alignment: A Mechanistic Perspective from Safety Neurons](https://mlsanity.com/paper/2406.14144/) (2024)
- [Understanding and Mitigating Numerical Sources of Nondeterminism in LLM Inference](https://mlsanity.com/paper/2506.09501/) (2025)
- [Constitutional Classifiers++: Efficient Production-Grade Defenses against Universal Jailbreaks](https://mlsanity.com/paper/2601.04603/) (2026)
- [LLMScan: Causal Scan for LLM Misbehavior Detection](https://mlsanity.com/paper/2410.16638/) (2024)
- [Probing Latent Subspaces in LLM for AI Security: Identifying and Manipulating Adversarial States](https://mlsanity.com/paper/2503.09066/) (2025)
- [DeepContext: Stateful Real-Time Detection of Multi-Turn Adversarial Intent Drift in LLMs](https://mlsanity.com/paper/2602.16935/) (2026)
- [Harmful Intent as a Geometrically Recoverable Feature of LLM Residual Streams](https://mlsanity.com/paper/2604.18901/) (2026)
- [Before the Last Token: Diagnosing Final-Token Safety Probe Failures](https://mlsanity.com/paper/2605.12726/) (2026)

## Built on by (most cited first, 2 of 2)

- [Probing the Prefill: Detecting Code Vulnerabilities via Latent Activations](https://mlsanity.com/paper/2608.16970/) (2026)
- [Same Probe, Different Numbers: Are Activation Probes Robust to Inference-Time Numerical Non-Determinism?](https://mlsanity.com/paper/2609.31796/) (2026)

## Abstract (arXiv)

Khatri et al. (2026) [DOI: 10.1109/DSN-W70714.2026.00027] show that lightweight MLP probes on final-layer activations of a single 8B model (LLaMA-3.1-8B) detect harmful prompts at F1 competitive with guard models 1000x larger, using one probe per benchmark. We reproduce this pipeline end-to-end and extend it along two axes the original study leaves open. First, we test whether the result generalizes across other model architecture and scale by training identical probes on activations from models like Gemma-4-E4B, Mistral-7B-v0.3, and Qwen2-7B, using the three benchmarks (WildJailbreak, BeaverTails, AEGIS 2.0). Second, we test how much of the reported performance is affected by non-determinism during inference by repeating extraction under five random seeds and measuring the variance of F1 scores. Our results reproduce the original LLaMA model benchmarks within 0.37 percentage points of the original F1 scores (and within 0.2 points on BeaverTails). We find that the original MLP probe architecture extends to other model families with F1 scores within a point of the values reported for LLaMA-3.1-8B. Our experiments varying seed values reveal an interesting observation: final token latent vectors remained the same for all tested architectures irrespective of the seed values used.