Safety · Concept
LLM Lie Detection and Deception Auditing
Methods for auditing when models hold knowledge they will not report, including internal recognition probes under concealment.
Forecast horizon
Verified model organisms and causal deception testbeds needed before lie-detection can be used in production auditing.
Connections
Connections · 5
How this node ties into the rest of the map, and the evidence behind each link.
from · constrains 5/10
CoT-Hidden Backdoors via Model Poisoning Benign-looking poisoned CoT undermines monitors that trust reasoning traces as action evidence.
+5 growth to · applies to 3/10
Emergent Misalignment from Narrow Fine-Tuning Lie detection methods are needed to audit models that may develop deceptive behaviors through fine-tuning or emergent misalignment.
+3 growth from · constrains 3/10
Invisible Reasoning via Filler Tokens in LLMs Filler-token invisible reasoning shows consequential computation can leave no interpretable CoT for monitors.
+3 growth from · informs 3/10
Scene-Level Audit of LLM-Generated Autobiography Confabulation Scene-level memoir audit quantifies grounded confabulation as a dominant failure mode for life-writing LLMs.
+3 growth from · informs 3/10
In-Context Neurofeedback: Privileged Access Test for LLM Self-Control Failure of privileged internal control weakens claims that LLMs have robust metacognitive access usable for safety.
+3 growthSignal sources
Signal sources
Dated facts from primary sources in this direction.
US evaluation centre Jun 2025
In June 2025 the US AI Safety Institute was renamed the Center for AI Standards and Innovation (CAISI), pivoting toward security, standards and adversary-model assessment.
NIST → Frontier safeguards May 2025
Anthropic activated its ASL-3 deployment and security standard with Claude Opus 4 on 22 May 2025 — the first real-world trigger of a responsible-scaling tier, focused on blocking bio-weapon uplift.
Anthropic → Cross-border testing 2025
The International Network of AI Safety Institutes (launched Nov 2024) ran a third joint testing exercise focused on agentic AI systems across cyber and fraud strands.
European Commission — AI Office →