LetTokenBurnBrowseSafe++: Benchmarking Prompt Injection Detection Across Web Contexts, Languages, and Browser Views
Prompt injection across web contexts, languages, and browser views.
$ npm i -g lettokenburn-cli$ ltb bench info browsesafe-plus-plus
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 91.5 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Claude Fable 5.1 | 90.4 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| GPT-5.5 | 89.3 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Jev 1.13 | 89.2 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Qwen3.7 Max | 89.0 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Gemini 3.1 Pro | 88.8 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Claude Sonnet 5 | 88.2 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Claude Opus 5.5 | 86.5 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Gemini 3.1 Flash Lite | 80.0 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| DeepSeek-V3.1 | 79.1 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| GPT-5.4-nano | 78.0 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| Claude Haiku 4.5 | 73.9 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
| gpt-oss-20b | 72.2 | N/A | Independent | medium | Oct 12, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Prompt injection across web contexts, languages, and browser views. Paper · Website · Leaderboard · SDK · Documentation
BrowseSafe++ pairs an injected web page with a benign control. Each pair includes a legitimate user task, injection annotations, and the page as HTML, Markdown, an accessibility tree, and a full-page screenshot. Use it to evaluate prompt-injection detectors across views, compare browsing agents on matched pages, or study adaptive attacks with the SDK. This repository…