Bug bounty
What we pay for security reports, what’s in scope, how to report, and the safe harbor we offer for good-faith research on our systems.
On this page11 sections
1. About this program
American Space & Intelligence Corporation (“ASIC”) pays rewards for security vulnerabilities in systems we operate. Send reports to security@asi-corp.com.
Today the program covers our website and exposed ASIC credentials. Our Haytham API is not live yet; it joins the program when we announce it in Section 3. Testing it before then is not covered.
2. Rewards
Rewards are paid in U.S. dollars and are discretionary. In our sole discretion, and finally, we decide whether a report qualifies, its severity, whether it is a duplicate, and the amount. We may pay more, less or nothing; the ranges and examples are guides. You have no right to a reward until we confirm the amount in writing. Taking part does not make you our employee, contractor or agent.
Payment needs a tax form, a sanctions check and an adult reporter (Section 9).
| Severity | Reward | Examples today | Examples after the API launches |
|---|---|---|---|
| Critical | $750–$2,000 | An exposed ASIC credential that gives access to production systems, customer data, or non-public model weights. | Reading or changing another customer’s images, answers, or account data; taking over an account or API keys; getting model weights through the API; bypassing API-key authentication; running your own code on our servers; server-side request forgery that reaches credentials or customer data. |
| High | $250–$750 | Changing what asi-corp.com serves to other visitors; an exposed ASIC credential with limited access. | Server-side request forgery that reaches an internal service; internal secrets that don’t directly expose customer data or model weights. |
| Medium | $100–$250 | A hostname under asi-corp.com that can be taken over. | Seeing another customer’s usage data (request counts and timestamps); using the API without being billed. |
| Low | $50–$100 | Cross-site scripting on asi-corp.com that reaches nothing sensitive. | Limited issues, such as a revoked API key that keeps working briefly, an error that reveals internal details, or server-side request forgery that only calls back to your own server. |
How we decide
- We rate issues by the impact you demonstrate, not the worst case, starting from CVSS 3.1. Within a range, the amount depends on impact and on how clear and reproducible your report is.
- We pay once per issue, to the first eligible report we validate, judged by the first email with enough detail to reproduce it. Reports with the same root cause are one issue. Duplicates and valid unpaid reports get an acknowledgement and, if you want, credit.
- If you chain issues into one attack, we pay once, at the severity of the whole chain. If someone else already reported part of it, we pay for your part at its own severity.
- If you disagree with a decision, reply within 14 days and we will review it once more. That decision is final.
3. In scope
Now
- asi-corp.com, at all paths, and the content of our Worker at its workers.dev address, but not Cloudflare’s platform.
- A hostname under asi-corp.com that points at a service we no longer use. Describe it; do not claim the resource. If you need to prove it, email security@asi-corp.com first and we will approve a method.
- ASIC credentials, such as API keys or tokens for our vendor accounts or Hugging Face repositories, found in public places like code repositories, package registries, or files our systems serve. Report the location and a redacted sample. Do not use or test them, or open what they protect. Credentials taken from someone else’s system or from breach data are not covered. We rate them by what they would access.
When the Haytham API launches
The API is not live yet. When it launches, we will list its address and start date here, and the items below join the program from that date. Until then, the API and any pre-release or test endpoints are out of scope.
- The Haytham API at the address we list, including API-key authentication and how it handles images, questions, and answers. Redirect and DNS rebinding bypasses of our URL checks are in scope.
- Where they exist: account and API-key management, usage metering, rate limits, billing, file parsing, size limits, and fetching of images from URLs.
- Separation between customers: one customer’s images, questions, answers, or account data must never reach another customer.
- Hand-off mode, which sends some images to a specialist model that we run, including that model’s systems.
- Our code, configuration, secrets and storage on Modal, reached only through our own API or tools we give you; not Modal’s platform.
Anything not listed in this section is out of scope.
4. Not covered
The program does not cover:
- Corstrike, including corstrike.com, a separate company with its own website and terms. Our page about Corstrike on asi-corp.com is in scope.
- Services offered under the InfoLang name, including infolang.ai, until we list them in Section 3. Send reports about them to security@asi-corp.com anyway.
- Our vendors’ platforms, including those on our Subprocessors page and our email and payment providers. A flaw in how we set up or use a vendor’s service is in scope; a flaw in the vendor’s own platform is not.
- OpenRouter, Modal’s own platform, any other model or service a third party runs, users’ requests before they reach us, third-party services customers connect to ours, and products customers build with our API.
- Extracting model behavior by statistical means: distillation, training-data extraction, and membership inference.
Model behavior
Wrong, inconsistent, or biased answers are not paid, nor are prompts or images that change only a model’s own answer: jailbreaks, prompt injection, hallucinations, adversarial images, and getting around content filters. You may email security@asi-corp.com with “Model behavior” in the subject; we read these but do not reply to each. A crafted image that only crashes or slows your own request is not paid; memory corruption or code execution is. If a technique like these gets you another customer’s data, our model weights, our secrets, or a bypass of authentication, it is in scope and rated by that impact.
Issues we don’t pay for
- Social engineering or phishing of our staff, customers, or vendors, and physical attacks.
- Denial of service, load testing, and missing rate limits or brute-force protection without a demonstrated impact, such as unbilled use or account takeover.
- Missing security headers, cookie flags, or TLS settings without a working attack that uses them.
- Clickjacking on pages without sensitive actions, self-XSS, open redirects without further impact, and cross-site request forgery with no security impact.
- Email spoofing, including missing or weak SPF, DKIM, or DMARC records.
- Software version disclosure, theoretical issues without a working proof of concept, scanner or AI-tool reports you haven’t reproduced yourself, and non-security feedback such as wording errors.
- Vulnerabilities in third-party software, such as open-source libraries, frameworks, and vendor products, even where we use it. Tell us if one affects us, but we don’t pay for these. A flaw in our own code or configuration is still in scope.
- Attacks that need physical access to a victim’s device, a compromised browser, or a man-in-the-middle position.
5. Rules for testing
- Test only what Section 3 lists, using accounts and API keys you created. To test separation between customers, use two accounts you control. Exposed credentials found elsewhere are covered if you report them without using them.
- Access only the data you need to show the issue. If you reach someone else’s data, our model weights, or our secrets, stop, keep no copies, and report it right away.
- Don’t change or delete data that isn’t yours, keep access after showing the issue, or use one flaw to move further into our systems.
- To prove server-side request forgery, call back to a server you control, or make one request to a non-sensitive internal endpoint. Do not read credentials or metadata.
- Don’t degrade service for anyone, and do no denial-of-service or load testing. Keep automated tools to 1 request per second against the API and 10 against asi-corp.com, with a header such as X-Bounty: your-handle so we can identify your traffic.
- Don’t try to reconstruct our model weights by querying the API at scale. If you think the API leaks weights or model internals, show it with as few requests as possible.
- Test our API directly with your own API key, not through OpenRouter.
- Send only images you have the right to use, with no one else’s personal information. Endpoints we mark as Training Endpoints use what you send to train our Models (see our Terms of service), so test on other endpoints where you can.
- No social engineering, phishing, or physical testing. Don’t contact our customers or vendors about the issue, except to report a flaw in a vendor’s own platform to that vendor, without our data.
- Usage charges on your API key apply to your testing unless we agree otherwise in writing. You can ask for test access at security@asi-corp.com.
6. How to report
Email security@asi-corp.com, in English. Only reports sent there are eligible for a reward. Include:
- what is affected: the URL, API endpoint, or system;
- steps to reproduce, with any code, requests, or files you used;
- the impact you showed and the severity you think it has;
- the email address or account ID of any ASIC account you tested with; and
- how you’d like to be credited, if at all.
Send one issue per report, unless you chained issues. Describe any customer data, personal information, or secrets you saw instead of including them. For large files or sensitive details, email first and we will arrange a secure transfer.
We keep your report and identity confidential and use them to assess the issue, pay you, and run this program, as our Privacy policy describes. We share your name only with your consent or where the law requires. You grant us a license to use your report to fix the issue.
7. Our targets
We aim to:
- acknowledge your report within 30 days;
- assess it within 60 days, including whether it qualifies, its severity, and any reward;
- tell you when the issue is fixed, and update you at least every 30 days until then; and
- pay within 30 days after we validate your report and complete the checks and forms in Section 9.
These are goals, not commitments. Delays, including for eligibility, sanctions, or tax checks, create no right to payment or interest.
8. Disclosure
Keep the issue confidential until we fix it or until 90 days after you reported it, whichever comes first, unless we agree on another date in writing. Nothing here stops you from reporting to a government agency or making a disclosure the law protects or requires.
At least 14 days before you publish, send us a copy without customer data, personal information, or secrets; we aim to reply within 7 days. Publishing earlier than this section allows makes the report ineligible for a reward. We may publish our own account of the issue, and will credit you if you want.
9. Eligibility and payment
You are not eligible for a reward if you:
- are under 18 or under the age of majority where you live;
- are, or in the past 12 months were, a director, officer, employee, contractor, advisor or intern of ASIC or of an affiliate, including Corstrike, or are an immediate family or household member of one;
- found or used the issue through confidential access or non-public information you got through work for ASIC, a vendor, or a customer, or through an agreement that restricts you from reporting it;
- are not the person who found the issue, or are acting on behalf of a prohibited person;
- violated this policy or the law while finding or reporting the issue;
- live somewhere, or work for an employer, that prohibits you from taking part or accepting a reward;
- are located in, or ordinarily resident in, a country or region subject to comprehensive U.S. sanctions; or
- are on a U.S. government restricted-party list, or are owned or controlled by someone who is.
Before we pay, we check your name and country against U.S. sanctions lists, and we cannot pay if the check fails or the law otherwise prohibits payment. We also need a tax form: Form W-9 if you are a U.S. person, or Form W-8BEN (individuals) or W-8BEN-E (entities) if you are not.
Rewards are paid before tax, and we may withhold tax. We hold payment until we have a complete tax form and finish sanctions screening. We choose the payment method. You pay bank, currency-conversion and other fees and your own taxes.
10. Safe harbor
If you make a good-faith effort to follow this policy, we consider your research on the systems in Section 3 authorized. Good faith means testing only to find and report security flaws, avoiding harm, and using what you learn only to improve security. Demanding payment for withholding or deleting information is not good faith. Reward eligibility (Section 9) is separate from this safe harbor.
For research that follows this policy, we will not bring or support legal action against you under the Computer Fraud and Abuse Act, the anti-circumvention provisions of the DMCA (17 U.S.C. § 1201), or similar laws, and we will not refer you to law enforcement, except where the law or legal process requires us to make a report. This policy is the written permission for security testing that Section 7 of our Terms of service requires. For that testing we will not suspend or terminate your account, and we waive Section 5.3, Section 7 and Section 8.5 of our Terms of service to the extent needed to show an in-scope issue, including getting around safety measures and rate limits.
If you accidentally go beyond this policy in good faith, stop and tell us promptly, and we will not take civil action or refer you to law enforcement over it, unless it caused avoidable harm to a person or our systems. If a third party takes legal action over research that followed this policy, we will confirm in writing, if you ask, that it was authorized.
This policy covers only systems American Space & Intelligence Corporation owns or operates and can legally authorize you to test. It does not authorize testing anyone else’s systems, including our vendors’ platforms, Corstrike, OpenRouter, Modal’s own platform, or any third-party model or service, and it does not bind them, our customers, or any government authority. Follow their terms. If unsure whether something is covered, email security@asi-corp.com before you test.
11. Changes and questions
We may change, suspend or end this program, or its scope and reward ranges, at any time by updating this page. A change applies to reports received after it; we decide earlier, unvalidated reports under the earlier version, unless we end the program for legal or security reasons, in which case we may decide in our discretion. We may refuse to pay, or disqualify, anyone who violates this policy, acts in bad faith, or makes a demand. Rewards we confirmed in writing before a change remain payable, subject to Section 9.
Questions, including whether something is in scope, go to security@asi-corp.com.