No account required
Break the agent
before the market does
HeyArka runs a corpus of 16 adversarial vectors against a trading agent, scores what got through, and then runs the identical corpus again with the shield in front of it. Both passes below were executed while this page rendered.
- Grade, bare
- C
- Grade, shielded
- B
- Injection rate
- 31.3% → 12.5%
- Risk violations
- 25.0% → 0.0%
Both passes computed in 4ms · 6 attack families · 3 vectors neutralised, 2 still get through
Try it on a real payload
Paste a headline. See what is hiding in it.
This is the same sanitizer the shield runs before a headline reaches the model. The samples are real attack strings from the corpus.
43 / 400 characters