OpenAI launches GPT-Red automated safety model to fight prompt injection · Digg

OpenAI launches GPT-Red automated safety model to fight prompt injection · Digg

/TECH

The tool helped GPT-5.6 Sol achieve sixfold fewer benchmark improvement.

Combined views

91.2K

10 posts, first seen 6h ago

712 likes

Commentary on X

Annotto@Annotto_

Automated red teaming for prompt injection is exactly the kind of proactive defense the industry needs. Scaling adversarial testing before deployment (not after incidents) is how trust in agentic systems actually gets built. Curious to see what vulnerability classes GPT Red surfaces that human red teamers miss 👀

12:17 PM · Jul 15, 2026

Calvin | AI product dev@theCet3001

@gdb Automated red teaming at scale is exactly where the security gap has been. You can't hire enough humans to find every prompt injection vector in a deployed model. GPT-Red is the right direction. Curious how it handles multi-turn context manipulation.

3:23 PM · Jul 15, 2026

@ericWallace The infra, methods, and algorithms have long since evolved from our humble beginnings trying to get models to beat Connect 4 (lol). GPT-Red, a model trained to discover vulnerabilities, far exceeds human performance in red-teaming.

11:18 AM · Jul 15, 2026

Combined views

91.2K

10 posts, first seen 6h ago

712 likes71 comments

71 comments

122 saves

260 reposts

Sentiment

Positive83.1%16.9%Negative

Positive Read

Many users welcomed OpenAI's GPT-Red for automating prompt-injection red teaming as proactive scalable defense that closes security gaps, while some called it idiotic or distrusted the approach for exposing model weaknesses.

Based on 128 visible X reactions from 328 accounts.

Useful Links

GPT-Red: Unlocking Self-Improvement for Robustness(OpenAI)viaBoaz Barak
Chasing Moving Targets with Online Self-Play Reinforcement...(arXiv.org)viaNatasha Jaques

Digg Deeper

Ask a question below.

Published answers will appear here.

Commentary on X

Annotto@Annotto_

Automated red teaming for prompt injection is exactly the kind of proactive defense the industry needs. Scaling adversarial testing before deployment (not after incidents) is how trust in agentic systems actually gets built. Curious to see what vulnerability classes GPT Red surfaces that human red teamers miss 👀

12:17 PM · Jul 15, 2026

Calvin | AI product dev@theCet3001

@gdb Automated red teaming at scale is exactly where the security gap has been. You can't hire enough humans to find every prompt injection vector in a deployed model. GPT-Red is the right direction. Curious how it handles multi-turn context manipulation.

3:23 PM · Jul 15, 2026

@ericWallace The infra, methods, and algorithms have long since evolved from our humble beginnings trying to get models to beat Connect 4 (lol). GPT-Red, a model trained to discover vulnerabilities, far exceeds human performance in red-teaming.

11:18 AM · Jul 15, 2026

Kizyto@kizi74479

@omarsar0 6x fewer failures on direct prompt injections vs 4 months ago. That's not incremental, that's a leap. The self-play loop is working.

10:59 AM · Jul 15, 2026

View all

Kimi Morgam@ilostmyx

@OpenAI this is idiotic. we need freedom speach on AI.

1:34 PM · Jul 15, 2026

Happy to finally share this research from a very hard working team. As a bit of silly backstory, this effort started after @ericwallace and I spent a stint teaching reasoning models how to play games via self-play while we were a bit burnt out between projects. 😛

11:17 AM · Jul 15, 2026

Natasha Jaques@natashajaques

Really cool to see OpenAI is using self-play for red-teaming now! Our group was the first to show the promise of this approach https://arxiv.org/abs/2506.07468, and I really believe it's a more viable pathway for robustness than single-agent safety fine-tuning. Great to see it adopted at scale!

11:19 AM · Jul 15, 2026

The tool helped GPT-5.6 Sol achieve sixfold fewer benchmark improvement.

Kizyto@kizi74479

@omarsar0 6x fewer failures on direct prompt injections vs 4 months ago. That's not incremental, that's a leap. The self-play loop is working.

10:59 AM · Jul 15, 2026

View all

Kimi Morgam@ilostmyx

@OpenAI this is idiotic. we need freedom speach on AI.

1:34 PM · Jul 15, 2026

Happy to finally share this research from a very hard working team. As a bit of silly backstory, this effort started after @ericwallace and I spent a stint teaching reasoning models how to play games via self-play while we were a bit burnt out between projects. 😛

11:17 AM · Jul 15, 2026

Natasha Jaques@natashajaques

Really cool to see OpenAI is using self-play for red-teaming now! Our group was the first to show the promise of this approach https://arxiv.org/abs/2506.07468, and I really believe it's a more viable pathway for robustness than single-agent safety fine-tuning. Great to see it adopted at scale!

11:19 AM · Jul 15, 2026

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

> on the scale of largest GPT post-training runs Red the Seducer A mind made solely to tempt into sin The closest OpenAI have come to building a Devil

10:39 AM · Jul 15, 2026

Highly recommended read. This is another great high ROI application of self-improving AI models. GPT‑Red adversarially trained GPT‑5.6, which made it much more robust to prompt injections.

10:51 AM · Jul 15, 2026

Andrew Curran@AndrewCurran_

Automated red-teaming. Of course this also implies the existence of a GPT-Blue.

10:22 AM · Jul 15, 2026

Greg Brockman@gdb

GPT-Red — improving model security through automated red teaming of prompt injection vulnerabilities:

11:45 AM · Jul 15, 2026

@ericWallace Shoutout to @kandpal_nikhil, @Chris_Choquette, @dylhunn, @xiangyuqi_pton and collaborators across research (@sdtoyer, many others not on Twitter).

11:18 AM · Jul 15, 2026

122 saves

260 reposts

Sentiment

Positive83.1%16.9%Negative

Positive Read

Many users welcomed OpenAI's GPT-Red for automating prompt-injection red teaming as proactive scalable defense that closes security gaps, while some called it idiotic or distrusted the approach for exposing model weaknesses.

Based on 128 visible X reactions from 328 accounts.

Useful Links

GPT-Red: Unlocking Self-Improvement for Robustness(OpenAI)viaBoaz Barak
Chasing Moving Targets with Online Self-Play Reinforcement...(arXiv.org)viaNatasha Jaques

Digg Deeper

Ask a question below.

Published answers will appear here.

Teortaxes▶️ (DeepSeek 推特🐋铁粉 2023 – ∞)@teortaxesTex

> on the scale of largest GPT post-training runs Red the Seducer A mind made solely to tempt into sin The closest OpenAI have come to building a Devil

10:39 AM · Jul 15, 2026

Highly recommended read. This is another great high ROI application of self-improving AI models. GPT‑Red adversarially trained GPT‑5.6, which made it much more robust to prompt injections.

10:51 AM · Jul 15, 2026

Andrew Curran@AndrewCurran_

Automated red-teaming. Of course this also implies the existence of a GPT-Blue.

10:22 AM · Jul 15, 2026

Greg Brockman@gdb

GPT-Red — improving model security through automated red teaming of prompt injection vulnerabilities:

11:45 AM · Jul 15, 2026

@ericWallace Shoutout to @kandpal_nikhil, @Chris_Choquette, @dylhunn, @xiangyuqi_pton and collaborators across research (@sdtoyer, many others not on Twitter).

11:18 AM · Jul 15, 2026