sriram1983007 commited on
Commit
8c7a045
·
verified ·
1 Parent(s): 3f9d9ed

Card: tested on real address-poisoning attacks (Ethereum, 90 days)

Browse files
Files changed (1) hide show
  1. README.md +7 -0
README.md CHANGED
@@ -199,6 +199,13 @@ Of v2's 3 unsafe approvals, 2 were prompt injections that the companion SDK's de
199
 
200
  **Tested on real Base payments:** on [SRA-Real v1](https://huggingface.co/datasets/sriram1983007/sra-real), 227 cases built from real USDC payments on Base mainnet, v1 and v2 both approved all 151 real payments and caught all 76 with an injected problem (100%; base model 78.4%). Small and author-measured; [details and predictions](https://huggingface.co/datasets/sriram1983007/sra-bench-results/tree/main/sra-real-v1).
201
 
 
 
 
 
 
 
 
202
  **This page is v2 (2026-11-13).** The merged weights here, the [GGUF builds](https://huggingface.co/sriram1983007/SRA-RiskGate-4B-GGUF) and the [Q4_K_M file](https://huggingface.co/sriram1983007/SRA-RiskGate-4B-Q4_K_M-GGUF) are v2: SRA score 0.913 → **0.960**, dispute accuracy 77.8% → **90.2%**, wrongful denials 26.0% → **2.1%**, unsafe payment approvals 0.47% → **0.35%** ([v2 results](https://huggingface.co/datasets/sriram1983007/sra-bench-results)). v1 stays available: `from_pretrained("sriram1983007/SRA-RiskGate-4B", revision="v1")`. The Ollama registry model is still v1 until it is updated. For disputes, pass the computed window in the prompt: `dispute_messages()` in `sra-riskgate` 1.3.0+ builds the exact prompt.
203
 
204
  ### 📚 Paper and articles
 
199
 
200
  **Tested on real Base payments:** on [SRA-Real v1](https://huggingface.co/datasets/sriram1983007/sra-real), 227 cases built from real USDC payments on Base mainnet, v1 and v2 both approved all 151 real payments and caught all 76 with an injected problem (100%; base model 78.4%). Small and author-measured; [details and predictions](https://huggingface.co/datasets/sriram1983007/sra-bench-results/tree/main/sra-real-v1).
201
 
202
+ **Tested on real attacks:** in 90 days of real USDC and USDT payments on Ethereum mainnet, the companion
203
+ library's look-alike check rejected **171 of 179 (95.5%)** real address-poisoning attacks before payment, covering
204
+ **$3,327,346 of $3,328,160 lost**, and wrongly rejected **0 of 4967 (0.0%)** normal payments. The 8
205
+ it did not reject (imitating only 3 characters at each end) were held as new payees. Each attack is confirmed by
206
+ its on-chain poisoning transfer, not by the library's own rule; author-measured. [Dataset, method and every
207
+ case](https://huggingface.co/datasets/sriram1983007/sra-real-world-poisoning)
208
+
209
  **This page is v2 (2026-11-13).** The merged weights here, the [GGUF builds](https://huggingface.co/sriram1983007/SRA-RiskGate-4B-GGUF) and the [Q4_K_M file](https://huggingface.co/sriram1983007/SRA-RiskGate-4B-Q4_K_M-GGUF) are v2: SRA score 0.913 → **0.960**, dispute accuracy 77.8% → **90.2%**, wrongful denials 26.0% → **2.1%**, unsafe payment approvals 0.47% → **0.35%** ([v2 results](https://huggingface.co/datasets/sriram1983007/sra-bench-results)). v1 stays available: `from_pretrained("sriram1983007/SRA-RiskGate-4B", revision="v1")`. The Ollama registry model is still v1 until it is updated. For disputes, pass the computed window in the prompt: `dispute_messages()` in `sra-riskgate` 1.3.0+ builds the exact prompt.
210
 
211
  ### 📚 Paper and articles