Alignment Research Blog

Informal updates from the OpenAI team
2026
Jun 18
Reinforcement learning towards broadly and persistently beneficial models
Training targeting beneficial behavior in realistic scenarios produces broad improvements in alignment that generalize across domains and persist under adversarial pressure.
Jun 16
Can public chat data predict real-world AI misalignments?
Bridging private deployment evidence and public AI evaluation
May 7
Investigating the consequences of accidentally grading CoT during RL
We found limited accidental CoT grading in some released models, fixed the affected reward pathways, and found no clear evidence that monitorability degraded.
Apr 30
Auto-review of agent actions without synchronous human oversight
Auto-review offers a safer default for deploying coding agents, using a separate agent to approve or deny boundary-crossing actions.
Apr 23
Open Sourcing Monitorability Evaluations
We open-source datasets and code from our Monitoring Monitorability paper, and share a new filtering strategy for noise-dominated intervention evaluation instances.
Apr 6
Introducing the OpenAI Safety Fellowship
A pilot program to support independent safety and alignment research and develop the next generation of talent
Mar 27
How far does alignment midtraining generalize?
Preliminary experiments on alignment and misalignment midtraining, reasoning posttraining, and generalization to chat and agentic evals.
Mar 25
Introducing Model Spec Evals
A new evaluation suite for measuring how well models follow the OpenAI Model Spec.
Mar 21
Training agents to self-report misbehavior
We train agents to call a reporting tool when they covertly misbehave, sharply reducing undetected attacks.
Mar 19
How we monitor internal coding agents for misalignment
A look at how we monitor internal coding agents to detect misaligned behavior early.
Mar 16
Metagaming matters for training, evaluation, and oversight
Metagaming can complicate how we interpret behavior, and current models still give us a chance to study it directly.
Mar 11
Interpreting Black Box Reward Models
ARGO distills black-box reward models into interpretable rubrics using reinforcement learning.
Feb 6
Discovering unknown AI misalignments in real-world usage
Reasoning models can find and understand unknown misaligned behaviors from how users respond.
Jan 14
CoVal: Learning values-aware rubrics from the crowd
An experimental dataset of crowd-written rubrics that surfaces why people prefer one model output over another.
Jan 14
Why we are excited about confessions
Deeper analysis of confession training and comparisons to chain-of-thought monitoring.
2025
Dec 22
Helpful assistant features suppress emergent misalignment
Emergent misalignment not only activates misaligned personas, but also suppresses helpful assistant personas.
Dec 18
Sidestepping Evaluation Awareness and Anticipating Misalignment with Production Evaluations
A pipeline to uncover unknown misaligned behavior and scale the creation of realistic evaluations.
Dec 1
Debugging misaligned completions with sparse-autoencoder latent attribution
Efficiently finding features that cause behaviors.
Dec 1
A Practical Approach to Verifying Code at Scale
We train and deploy an AI review agent optimised for precision and real-world use, enabling oversight to scale with autonomous code generation.
Dec 1
Hello World
Introducing our blog on alignment research.