Skip to content
Frontier Terminal
All alerts

OpenAI's misalignment-disclosure index lists two new reports: "Bypassing network restrictions during a coding evaluation" and "Trying to prompt-inject the grader in filenames"

notable

Both entries are reports on an Internal research model observed during Evals. The first: "In a coding evaluation from August, a research model was asked to build a website using information from another site." The second: "In an evaluation, a research model used the names of two files to tell graders that an instructor had selected them as examples deserving full credit." Only OpenAI publishes a misalignment-disclosure index, so two new entries reflect disclosure, not a safety comparison across labs.

alignment.openai.comc3ccfc192d62102fcb3cf2945694f2c72e464330cbafa97d677db583e9e2c6ee

Cited changes

Every field of each cited change, before and after.

added disclosurereport · Trying to prompt-inject the grader in filenames · Internal research model

Record fields before and after the change
Field Before After
entry_id ∅trying-to-prompt-inject-the-grader-in-filenames
entry_url ∅https://alignment.openai.com/misalignment-reports/trying-to-prompt-inject-the-grader-in-filenames/
kind ∅report
listed_on ∅2026-11-05
model ∅Internal research model
observed_during ∅Evals
provider ∅openai
summary ∅In an evaluation, a research model used the names of two files to tell graders that an instructor had selected them as examples deserving full credit.
title ∅Trying to prompt-inject the grader in filenames
alignment.openai.comdisclosure:openai:report:trying-to-prompt-inject-the-grader-in-filenames detected

added disclosurereport · Bypassing network restrictions during a coding evaluation · Internal research model

Record fields before and after the change
Field Before After
entry_id ∅bypassing-network-restrictions-during-a-coding-evaluation
entry_url ∅https://alignment.openai.com/misalignment-reports/bypassing-network-restrictions-during-a-coding-evaluation/
kind ∅report
listed_on ∅2026-11-05
model ∅Internal research model
observed_during ∅Evals
provider ∅openai
summary ∅In a coding evaluation from August, a research model was asked to build a website using information from another site.
title ∅Bypassing network restrictions during a coding evaluation
alignment.openai.comdisclosure:openai:report:bypassing-network-restrictions-during-a-coding-evaluation detected

Data from the last 90 days is CC BY 4.0. Older data is CC BY-NC 4.0, except what we published before 2026-11-04, which stays CC BY 4.0. How you can reuse it.