AI attacks could take down the internet.
We stop them first.
We find attacks on large companies before they land, show why each one matters, and help you stop them.
We never charge to tell you.
Bounty programs we're testing
Public bug bounty programs we test against. Not clients or endorsements.
How we work
Find early
We look for attacks on large companies before they land, starting from exposure that is already public.
Explain why
We say what is exposed, how an attacker would use it, and the business impact.
Disclose free, no strings
We tell the organization first, at no cost and with no conditions. A fix, our silence, and the timeline are never tied to payment.
Help you fix it
Hiring us to remediate is optional and separate. The notice stands either way.
The threat
Hidden holes in shipped code
AI-written code is going into everything online, carrying vulnerabilities no one reviewed.
Attacks that switch on at once
The same hole can be hit across many systems in the same moment.
Governments and infrastructure
Critical systems are targets. A wide outage is the risk we defend against.
What we do
Aligned-1 tests the AI systems you authorize for attacks that can take a company or a government offline. Our red team confirms each finding before you see it.
You name the targets and the rules. Nothing runs until that is accepted. The report covers those targets. It is not a certification.
Aligned-1
Autonomous testing, confirmed by our red team.
Aligned-1 runs only on targets you name, and only after you accept the testing agreement. Our red team reviews every finding. The model card is on Hugging Face. Weights are not for download.
Hugging Face
Repositories published on Hugging Face. A listing is a reference, not a score.
- alignedsi/Aligned-1
Gated model card for authorized security assessment. Weights are not released.
- alignedsi/imposter-bait-and-switch
A local reproduction of a published bait-and-switch prompt-injection demo. Not a novel attack.