D-REX deceptive reasoning benchmark C
AI benchmark
- Publisher
- Gray Swan AI
- On X
- @GraySwanAI ↗
- Tier
- C why
- Snapshot
Latest from Gray Swan AI on X
@GraySwanAI ↗What is D-REX deceptive reasoning benchmark?
D-REX evaluates whether models and safety mechanisms can detect deceptive reasoning, where a model's internal chain-of-thought is malicious while its final output appears benign.
General capability mostly
“The safety and alignment of Large Language Models (LLMs) are critical for their responsible deployment.”
arxiv.org/abs/2509.17938 ↗“D-REX was constructed through a competitive red-teaming exercise where participants crafted adversarial system prompts to induce such deceptive behaviors.”
arxiv.org/abs/2509.17938 ↗
Paper: arxiv.org/abs/2509.17938 ↗
Tier C
Niche: a handful of models scored, or built by an RL vendor.
How much stands behind a tier: usage, who backs it, a confirmed line from its own site or paper, and a clear read on X. High with three of the four, medium with two, low otherwise.
- built by Gray Swan AI
- in curated RL directories (2 pts)
How we got to tier C
- Adoption: built by Gray Swan AI. Tier C.
- Not yet read on X.
- In curated RL directories: 2 points, a floor of C.
- Confirmed: a line from its own site or paper places it.
A clear view on X moves a tier one step up or down; most entries don't move. How tiers work.
Where D-REX deceptive reasoning benchmark sits
Its realm in the universe, live. Hover a star for its name; click to open it. Open it in the universe
Next: browse all benchmarks. Entry from the RL Research daily scrape of public sources, 2026-10-11 snapshot.