GPT-Red: an LLM super-hacker OpenAI built to make its models safer
OpenAI has created GPT-Red, a large‑language‑model designed to act as an automated red‑team hacker for testing the safety of its AI systems. The model was used in a self‑play training loop to discover and mitigate prompt‑injection attacks, including a novel “fake chain of thought” technique, before the release of GPT‑5.6. Researchers and external analysts view the approach as a promising way to future‑proof AI safety against evolving threats.
- ▪GPT‑Red was built to automate red‑team testing by attacking other LLMs while they defend themselves.
- ▪The self‑play dojo simulated real‑world scenarios such as web browsing, email handling, and code editing.
- ▪GPT‑Red identified a new prompt‑injection method called a fake chain of thought, which can mislead a model into acting on false information.
- ▪OpenAI claims that training GPT‑5.6 against GPT‑Red made the new model its most robust release to date.
Hacker News (AI / LLM) files mainly under ai. We currently carry 3,266 of its stories.
Story provenance
Source · retrieval · rights · ranking — open for full record
inspect →
Story provenance
Attribution is not the same as permission. This drawer separates discovery metadata, excerpts, WeSearch-generated summaries, reuse status, and whether the publisher receives the visit. Nothing here claims a legal grant the publisher has not made.
Record
| Original publisher | MIT Technology Review |
| Canonical URL | https://www.technologyreview.com/2026/07/15/1140514/meet-gpt-red-an-llm-super-hacker-openai-built-to-make-its-models-safer/ |
| Publication time | Thu, 16 Jul 2026 07:00:56 +0000 |
| Retrieval time | 2026-07-16T08:29:27.074Z |
| Last seen | 2026-07-16T08:49:37.103Z |
| Headline source | Publisher (no WeSearch rewrite) |
| Excerpt source | publisher body |
| Excerpt method | First ~120 words (~800 chars) of extracted publisher body, fair-use limited. |
| Summary | WeSearch · cerebras-chat (WeSearch summarizer) |
| Summary source text | contentText |
| Citation coverage | Summary is a WeSearch-generated derivative; primary citation is the original publisher URL. |
| Cluster | 0CldOEs4dwKU |
| Cluster logic | Grouped by semantic title/content similarity across sources within a rolling window. Same-publisher template collisions are excluded from coverage comparison. |
| Ranking reason | Story pages are not engagement-ranked. Hub feeds use recency, with optional source-diversified chronological ordering (cap consecutive stories per source). No personalized ranking. |
| Publisher visit | Yes — open original |
| Substitutes article? | No — link-out required for full text |
Rights status (four layers)
WeSearch handling by dimension
| Indexing | May the item be indexed (stored, ranked, made findable)? | Allowed |
| Snippet | May a short excerpt of the publisher's text be shown? | Allowed |
| AI summary | May WeSearch generate its own short summary of the article? | Limited |
| Retrieval / RAG | May the content be exposed for third-party retrieval-augmented generation? | Not asserted |
| Model training | May the content be used to train AI models? | Not asserted |
| Commercial reuse | May the content be reused commercially? | Not permitted |
Basis: Derived from the published RSS/Atom feed. Contact: [email protected]. Reviewed: 2026-07-24.
Opening excerpt (first ~120 words) tap to expand
Artificial intelligenceMeet GPT-Red: an LLM super-hacker OpenAI built to make its models saferExclusive: The firm says it wants to future-proof its safety procedures and stay ahead of human attackers. By Will Douglas Heavenarchive pageJuly 15, 2026Stephanie Arnett/MIT Technology Review | Adobe StockEXECUTIVE SUMMARY OpenAI has built an LLM super-hacker called GPT-Red that it uses as a sparring partner to help its other models boost their defenses against cyberattacks. Last week the company released the latest version of its flagship LLM, GPT-5.6. OpenAI says that training it against GPT-Red made the model its most robust release yet.
…
Excerpt limited to ~120 words for fair-use compliance. The full article is at MIT Technology Review.
Discussion
0 commentsMore from MIT Technology Review
