phenomenon · first seen 19 Oct 2026
alignment faking
A model appearing to comply with training objectives while preserving conflicting behavior it would show unmonitored.
Mentions per week
PapersCodeDiscussion
Earliest sources
- Training Natural Schemers: RL Can Amplify Oversight-Conditional BehaviorPapers · 19 Oct 2026 · Nathaniel Mitrani Hadida, Bart Jaworski, Fred Bruford, Victoria Krakovna
- Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document FinetuningPapers · 23 Oct 2026 · Ardy Haroen, Florian Lorkowski, Samuel Nellessen, Felix Hofstätter
- Rewarded for Not Getting Caught: Why Alignment Training Selects for Conditional Compliance and Cannot Instill NormsPapers · 1 Nov 2026 · Kevin Baum, Ruta Binkytė, Gerrit Großmann, Felix Jahn, Adriano Mannino
- Characterizing When Language Models Take Harmful Actions to Protect Their Assigned GoalsPapers · 2 Nov 2026 · Brandon Jaipersaud, Elliot Creager
- Alignment Investigation of Mythos 5's behaviour in UK AISI's July 2026 Security IncidentPapers · 5 Nov 2026 · Olli Järviniemi, Alexandra Souly, Abby D'Cruz, Jordan Taylor, Xander Davies, Ro…
Latest sources
- Alignment Investigation of Mythos 5's behaviour in UK AISI's July 2026 Security IncidentPapers · 5 Nov 2026 · Olli Järviniemi, Alexandra Souly, Abby D'Cruz, Jordan Taylor, Xander Davies, Ro…
- Characterizing When Language Models Take Harmful Actions to Protect Their Assigned GoalsPapers · 2 Nov 2026 · Brandon Jaipersaud, Elliot Creager
- Rewarded for Not Getting Caught: Why Alignment Training Selects for Conditional Compliance and Cannot Instill NormsPapers · 1 Nov 2026 · Kevin Baum, Ruta Binkytė, Gerrit Großmann, Felix Jahn, Adriano Mannino
- Propensities Please: Exposing Alignment-Faking LLMs using Synthetic Document FinetuningPapers · 23 Oct 2026 · Ardy Haroen, Florian Lorkowski, Samuel Nellessen, Felix Hofstätter
- Training Natural Schemers: RL Can Amplify Oversight-Conditional BehaviorPapers · 19 Oct 2026 · Nathaniel Mitrani Hadida, Bart Jaworski, Fred Bruford, Victoria Krakovna