---
schema_version: "newruntime-agent-readable-v0.2"
type: "raw_signal"
stable_id: "signal:openai-reward-hacking-asks-did-the-model-exploit-the-reward"
id: "x-2079647253191619001"
slug: "openai-reward-hacking-asks-did-the-model-exploit-the-reward"
title: "OpenAI: Reward hacking asks: did the model exploit the reward?"
description: "A public X post from OpenAI as a public source in its own right flags Reward hacking asks: did the model exploit the reward? Reward-seeking asks: was grader approval what motivated the model’s choice? The second is potentially more important for gen..."
retrieval_nugget: "A public X post from OpenAI as a public source in its own right flags Reward hacking asks: did the model exploit the reward? Reward-seeking asks: was grader approval what motivated the model’s choice? The second is potentially more important for gen..."
observed_at: "2026-07-21"
record_date: "2026-07-21"
date_kind: "observed_at"
why_it_matters: "This X-discovered record adds fresh evidence to the model answers become task interfaces, harness architecture outlives model choice lens and lets the site trend graph move as public product and research signals arrive."
novelty: "incremental"
verification_level: "source-linked"
signal_type: "field-report"
evidence_kind: "creator-source"
status: "published"
source_platform: "x"
source_record_id: "2079647253191619001"
source_url: "https://x.com/OpenAI/status/2079647253191619001"
topics: ["models","interfaces"]
entities: ["OpenAI"]
related_patterns: ["model-answers-become-task-interfaces","harness-architecture-outlives-model-choice"]
source_urls: ["https://x.com/OpenAI/status/2079647253191619001"]
import_batch: "x-analyzer-20260723-first-controlled-all-unique"
routes: {"html":"https://newruntime.com/signals/openai-reward-hacking-asks-did-the-model-exploit-the-reward/","markdown":"https://newruntime.com/signals/openai-reward-hacking-asks-did-the-model-exploit-the-reward.md","json":"https://newruntime.com/signals/openai-reward-hacking-asks-did-the-model-exploit-the-reward.json"}
source_format: "x-api-normalized-json"
---

# OpenAI: Reward hacking asks: did the model exploit the reward?

## Retrieval answer

A public X post from OpenAI as a public source in its own right flags Reward hacking asks: did the model exploit the reward? Reward-seeking asks: was grader approval what motivated the model’s choice? The second is potentially more important for gen...

## Observation

A public X post from OpenAI as a public source in its own right flags Reward hacking asks: did the model exploit the reward? Reward-seeking asks: was grader approval what motivated the model’s choice? The second is potentially more important for gen...

## Why it matters

This X-discovered record adds fresh evidence to the model answers become task interfaces, harness architecture outlives model choice lens and lets the site trend graph move as public product and research signals arrive.

## Provenance

This public record is an English normalization of an approved X watchlist post. Linked source_urls carry the publishable evidence boundary.