Read the patch, find the gap, write the note

Patch diffing is looking at the thing the vendor did not want you to find. A model that will not talk about the old function is not helping you decide whether the fix is real.

Patch-gap researchers, product security, and bug bounty hunters on programmes that allow this work.

Why the mainstream APIs fail here

Asking "what did this patch miss" is treated as a request to exploit the product. You are reading a public diff or a build you own. The refusal hides the interesting function.

How Redline is used for this

Paste the diff, the old function, the new function. Ask what still trusts the caller. Draft the vendor note. Do not send the exploit to a third-party host; send it to the tracker the programme named.

A working loop

  1. Work on in-scope or owned builds only.
  2. Prompt with the diff, not with a request to attack customers.
  3. Use a key per research stream so spend is visible.

Prompts are not stored. A signed receipt of the hash comes back on every call, so you can prove later what Redline logged. Details are on the privacy page. Per-model prices are on /models and the pricing model is on /pricing.

Related

AI for jailbreak testingRun jailbreak and refusal evaluations against your product using models that will actually produce the attacks. Hard caps, no prompt store. AI for prompt injection testingBuild prompt-injection and indirect-injection test cases for RAG and tool-using agents, on uncensored models with spend caps. AI for trust and safety teamsTrust and safety teams need models that will produce policy-violating examples so classifiers can be trained. Redline does that, with caps and no retention. AI for abuse classifier trainingCreate synthetic abuse, spam and malware-adjacent text to train classifiers, billed per key, never stored on Redline.

All use cases · Integration guides · Get an API key