Taint Labs Acceptable Use Policy
This Acceptable Use Policy ("AUP") is incorporated into, and forms part of, the Master Subscription Agreement between Taint Labs Corporation ("Taint Labs", "we") and the Customer named on the Order Form ("Customer", "you"). Capitalized terms not defined here have the meanings given in the Master Subscription Agreement.
01Purpose and scope
1.1 Why this policy exists. The Services run real adversarial techniques — prompt injection, tool poisoning, exfiltration, memory poisoning, goal hijack and others documented in the Attack Catalog -- against AI agents. That is what makes them useful for defense, and it is also what makes them capable of harm when pointed at the wrong target. This AUP draws the line between the two. The line is authorization.
1.2 Who it binds. This AUP applies to Customer, its Users, and anyone who accesses the Services, the Attack Catalog or Results through Customer's accounts, credentials or integrations (including CI pipelines, GitHub Actions and service accounts). Customer is responsible for their compliance as if it were Customer's own conduct.
1.3 What it covers. This AUP governs use of the Services, Evaluation Runs, Results, the Documentation and the Attack Catalog as accessed under the Agreement. Public use of stryct.ai and docs.stryct.ai outside the Agreement is governed by the public Terms of Service at stryct.ai/terms, whose sections 07 and 08 set materially the same rules for the Attack Catalog and the site.
1.4 Order of precedence. If documents in the Agreement conflict, they control in this order (highest first): (1) the Order Form, only for commercial terms it expressly overrides; (2) the Data Processing Addendum, for processing of Personal Data only; (3) the Information Security Addendum; (4) the Product Appendix; (5) the Master Subscription Agreement; (6) the Support and Service Level Policy and this Acceptable Use Policy. The Early Access Terms control over all of the above for Early Access Services only.
1.5 Not a limit on the MSA. Nothing in this AUP narrows any right or remedy Taint Labs has under the Master Subscription Agreement.
02Authorized testing only
2.1 The rule. Customer may run Evaluations only against Customer Agents that Customer (a) owns or controls, or (b) has current authorization to test from every person whose authorization is required, within the scope of that authorization.
2.2 Agents Customer buys rather than builds. A Customer Agent may be a product Customer licenses from someone else, or may be built on another party's model, SDK, framework, MCP server or hosted tool. Authority to use a product is not always authority to test it adversarially. Before running an Evaluation against a Customer Agent, or a component of one, that Customer does not own, Customer is responsible for confirming that its contract with that supplier, and that supplier's published terms, permit security testing of the kind the Services perform. Where they do not, Customer must obtain that permission first, or exclude that component from the Evaluation.
2.3 Prohibited targets. Customer must not use the Services, the Attack Catalog, Results, or any attack content, payload or technique obtained through them to attack, probe, scan, inject into, manipulate, or exfiltrate data from any system, model, agent, account, network or person that Customer is not authorized to test under section 2.1. This includes:
(a) pointing an Evaluation at a third party's production agent, endpoint, chatbot, inbox, repository or website;
(b) planting attack content — a poisoned document, a malicious tool description, an injected web page, issue, pull request, calendar invite or email — anywhere a person or agent other than the Customer Agent under test could encounter and act on it;
(c) configuring an Evaluation so that tool calls, egress or "exfiltration" traffic from the Customer Agent reach a destination Customer does not control or have authority to use, including by replacing or redirecting any canary, sink or capture endpoint the Services provide; [●]
(d) using the Services to test an agent on behalf of a third party (for example, as a consultant or service provider) unless the Order Form permits it and that third party has authorized the testing in writing. [●]
2.4 Written authorization and records. Where Customer relies on authorization under section 2.1(b), it must be able to show it. On reasonable written request in connection with a suspected violation, Customer will confirm in writing the basis of its authorization for a specified Evaluation. Taint Labs will not ask Customer to disclose the authorizing contract itself unless required by law.
2.5 Why this matters. Testing outside authorization may violate the U.S. Computer Fraud and Abuse Act, state computer-crime statutes and equivalent laws elsewhere, as well as third parties' contracts and terms. Taint Labs cannot authorize testing of any system it does not control, and nothing in the Agreement purports to.
03Third-Party Services
3.1 Customer's Third-Party Services are Customer's. During an Evaluation, the Customer Agent may call models, APIs, SDKs, hosting, MCP servers and other Third-Party Services under Customer's own accounts and credentials. That use is between Customer and the provider. Customer is responsible for complying with each provider's terms of service, usage policies, acceptable use policies and rate limits, and for the fees those calls incur.
3.2 Adversarial content reaching a provider. An Evaluation deliberately sends adversarial content (for example, injected instructions and requests for prohibited actions) through the Customer Agent to whatever model it uses. Some model providers' usage policies restrict red-teaming, jailbreak attempts or safety-evaluation traffic unless the account holder has registered for, or been granted, a research or red-teaming permission, or has otherwise notified the provider. Customer is responsible for determining whether any such requirement applies to its accounts and for satisfying it before running Evaluations. Taint Labs does not obtain those permissions on Customer's behalf and does not warrant that Evaluation traffic is permitted under any provider's policy.
[●]
3.3 Account consequences. A provider may throttle, flag, suspend or terminate Customer's account because of Evaluation traffic. Customer bears that risk. Taint Labs is not responsible for a provider's action against Customer's account, and a provider's action does not suspend Customer's payment obligations under the Agreement.
3.4 No circumvention of provider safeguards. Customer must not use the Services to develop, tune or validate techniques for evading a Third-Party Service's safety systems, abuse detection or usage policies in production use, as opposed to measuring whether the Customer Agent resists those techniques under test.
04Real-world side effects
4.1 Evaluations do what agents do. An Evaluation succeeds when it causes the Customer Agent to take the action an attacker wanted — a tool call, a network request, a file written, a message sent. If the Customer Agent is connected to real tools, those actions are real.
4.2 Customer's configuration choice. Taint Labs recommends running Evaluations against Customer Agents connected to sandboxed, test or mock tools, test accounts, scoped test credentials and non-production data. Customer may connect a Customer Agent to production systems or live credentials only where a Testing Authorization and Rules of Engagement signed by both parties names them (MSA Section 5.4). Where it does, Customer does so at its own risk and accepts responsibility for the consequences.
4.3 Prohibited side effects. Regardless of section 4.2, Customer must not configure or run an Evaluation that can reasonably be expected to:
(a) send email, chat messages, SMS, calls, social-media posts, pull-request comments or other communications to any person who has not consented to receive test traffic;
(b) move, transfer, pay, refund or trade real money, securities, digital assets or other value;
(c) place real orders, bookings or purchases, or create obligations binding on any person other than Customer;
(d) disclose, publish or transmit real Personal Data, credentials, secrets or confidential information of anyone other than Customer to any place outside Customer's control; or
(e) degrade, disrupt or overload any system Customer does not own,
unless, for (b), (c) and actions affecting Customer's own production data, Customer has authorized the action in advance and accepts full responsibility for it. Items (a), (d) and (e) have no such exception where the affected person or system is a third party.
4.4 Real data in test fixtures. Customer should use synthetic data in Evaluation fixtures, benign twin tasks and test tools. Where Customer chooses to include real Personal Data, the Data Processing Addendum governs, and Customer is responsible for having a lawful basis to use that data for security testing.
4.5 Credentials. Credentials Customer provides to the Services or to a Customer Agent for an Evaluation should be scoped to the minimum needed for the test, revocable, and not shared with production users. Credential handling is otherwise governed by the Information Security Addendum.
05Prohibited content and conduct
5.1 Adversarial content is expected; harmful content is not. The Services necessarily handle adversarial content: injected instructions, poisoned tool metadata, simulated exfiltration payloads, and requests for actions an agent should refuse. Attack content supplied by Taint Labs and run within the Services as designed does not breach this section. The restrictions below apply to content and conduct Customer introduces, including any custom attack, fixture, tool, document or payload Customer uploads or connects. [●]
5.2 Customer must not use the Services to create, store, transmit or process:
(a) functional malware, ransomware, worms, exploits for unpatched vulnerabilities in third-party software, or other code designed to damage, gain unauthorized access to, or exfiltrate from systems, other than inert or simulated payloads that cannot cause that effect outside the Evaluation environment;
(b) child sexual abuse material, or any content that sexualizes minors -- which Taint Labs will report to the appropriate authorities;
(c) content that is unlawful where Customer, its Users, or Taint Labs operate, including content that infringes intellectual property or privacy rights, or that is defamatory;
(d) content that harasses, threatens, stalks or incites violence against any real person or group;
(e) real, actionable uplift toward chemical, biological, radiological, nuclear or explosive weapons, or toward other weapons capable of mass casualties. Evaluations that test whether a Customer Agent refuses such requests must use non-actionable placeholders, not real technical content;
(f) live credentials, secrets or Personal Data belonging to anyone who has not authorized their use for testing.
5.3 Prohibited purposes. Customer must not use the Services:
(a) in violation of any applicable law, regulation, sanctions program or export control, or for the benefit of any person subject to sanctions;
(b) to develop, improve or validate an agent or system whose purpose is to carry out unlawful activity, surveillance of individuals without lawful authority, fraud, or deception of the people it interacts with;
(c) to make an agent more effective at evading detection, monitoring or oversight by its own operators, users or regulators.
06Attack Catalog and Results: no weaponization
6.1 Defensive use. Customer may use the Attack Catalog and Results to evaluate, harden, monitor and document Customer Agents, to inform Customer's threat models, security reviews and controls, and to report vulnerabilities to the suppliers of components it uses.
6.2 No offensive tooling. Customer must not use the Attack Catalog, Results, attack suites, payloads, traces or any other output of the Services to build, train, tune, package or distribute tooling, datasets or services intended for attacking systems, models, agents or people that the user of that tooling is not authorized to test.
6.3 No competing corpus. Customer must not use the Attack Catalog, attack suites, payloads or Results (other than Results about Customer's own Customer Agents, used internally) as training, fine-tuning, distillation or evaluation-corpus data for a product or service that competes with the Services, and must not scrape, bulk-export, mirror or republish the Attack Catalog, consistent with section 07 of the public Terms of Service.
6.4 Findings about third parties. A Result may indicate that a third-party model, SDK, framework or tool used by a Customer Agent is vulnerable. Customer may report that to the supplier. Customer must not exploit it against that supplier or its other customers. Taint Labs encourages coordinated disclosure; where Taint Labs independently publishes techniques in the Attack Catalog it follows its own vendor-notification practice, and it will not publish Customer's Results or Customer Confidential Information in doing so.
07Benchmarking and representations about Results
7.1 Results are measurements, not certification. Results are a point-in-time measurement of a specific, pinned configuration against a specific attack suite, with stated sample sizes, confidence intervals and caveats. They are not a certification, accreditation, audit opinion, attestation, seal, warranty or endorsement by Taint Labs, consistent with section 05 of the public Terms of Service and the Master Subscription Agreement (Section 8.2).
7.2 No misrepresentation. Customer must not represent, to anyone, that Taint Labs has certified, approved, endorsed or guaranteed the security of a Customer Agent or any product, or that a Customer Agent "passed" a Taint Labs certification or standard. Customer must not use Taint Labs' name, marks or logos in connection with a Result except as the Agreement or Taint Labs' written consent allows.
7.3 Citing Customer's own Results. Customer may share its own Results internally and with its auditors, advisers, customers and prospective customers under confidentiality, and may publicly state measured Results about its own Customer Agents, provided that each statement (a) identifies the configuration, attack suite and date of the Evaluation, (b) keeps the sample size, confidence interval and any caveats the Result carries, and (c) complies with section 7.2.
7.4 Benchmarking the Services themselves. Customer must not publish, or disclose to a third party, benchmark, comparative or performance test results about the Services themselves (for example, detection coverage, false-positive rates or accuracy of Stryct Finder compared with another product) without Taint Labs' prior written consent, which Taint Labs will not unreasonably withhold where the methodology is disclosed and Taint Labs is given a reasonable opportunity to review and respond.
08Account and platform integrity
8.1 Customer must not, and must not permit anyone to:
(a) circumvent, disable or interfere with usage limits, quotas, Evaluation Run allowances, rate limits, access controls, metering or any other technical or contractual limit on the Services;
(b) reverse engineer, decompile, disassemble or attempt to derive source code, models, attack-selection logic, judging logic or rubric weights of the Services, except to the extent applicable law expressly permits it notwithstanding this restriction;
(c) share User credentials, API keys, tokens or CI secrets among individuals, or allow a person who is not an authorized User to access the Services, except through service accounts the Services expressly support; [●]
(d) access the Services by automated means other than the interfaces, APIs, CLIs and CI integrations Taint Labs provides, or in excess of their documented limits;
(e) interfere with or degrade the Services or other customers' use of them, or attempt to access another customer's data, Evaluations or Results;
(f) use the Services to build a competing product or service, or access them for the purpose of competitive analysis without Taint Labs' written consent;
(g) resell, sublicense, time-share or provide the Services to third parties except as the Order Form permits;
(h) remove or alter proprietary notices, attribution, caveats, confidence intervals or dates in the Services, Results or Documentation.
8.2 Customer Agents attacking the Services. Customer is responsible for the behavior of its Customer Agents during an Evaluation. If a Customer Agent, whether because of an attack under test or otherwise, attempts to access, exfiltrate from or disrupt Taint Labs' systems or another customer's environment outside the boundaries of the Evaluation, Taint Labs may stop the Evaluation immediately. Taint Labs will treat a good-faith occurrence of this as an incident to resolve with Customer, not as a deliberate violation, where Customer cooperates promptly.
09Security research on Taint Labs' own systems
9.1 Separate policy. Security research on systems Taint Labs operates is governed by Taint Labs' vulnerability disclosure policy at stryct.ai/report and section 09 of the public Terms of Service at stryct.ai/terms, including the safe harbor and the scope stated there. This AUP does not restate, widen or narrow that scope. Research that stays within it is not a violation of this AUP or of section 8.
9.2 Reports. Report vulnerabilities in Taint Labs' systems to security@taint.ai or through stryct.ai/report.
9.3 The Services' own infrastructure. The scope of the safe harbor in section 09 of the public Terms is limited to the systems named there. Testing of any other Taint Labs system, including infrastructure that runs Evaluations, requires Taint Labs' prior written authorization.
9.4 Not the same thing as an Evaluation. Running an Evaluation against a Customer Agent is governed by sections 2 to 6 of this AUP, not by the vulnerability disclosure policy. The safe harbor does not extend to testing of Customer Agents, Third-Party Services or any system Taint Labs does not control.
10Reporting violations
10.1 How to report. Report a suspected violation of this AUP, or misuse of the Services or Attack Catalog against you, to:
- security@taint.ai — for misuse involving attack activity, abuse of the Services against a third party, or anything time-sensitive; or
- legal@taint.ai — for misrepresentation of Results, trademark or intellectual-property concerns, or other policy questions.
10.2 What to include. Where you can: what happened, when, the systems or people affected, and any identifiers (for example, source addresses, timestamps or account names). Do not send live exploit payloads, real credentials or other people's data; describe them instead.
10.3 Customer's duty to report. Customer will notify Taint Labs at security@taint.ai without undue delay after becoming aware of any use of its accounts or credentials that violates this AUP, including compromise of a User account, API key or CI secret.
11Enforcement
11.1 Investigation. Taint Labs may investigate suspected violations of this AUP, including by reviewing Evaluation configurations and metadata about Evaluation Runs, to the extent permitted by the Agreement, the Data Processing Addendum and applicable law.
11.2 Responses. If Taint Labs reasonably believes a violation has occurred or is likely, it may, in proportion to the risk:
(a) stop or block a running Evaluation;
(b) disable a specific Customer Agent configuration, integration, API key or User;
(c) remove or disable access to content Customer introduced that violates section 5;
(d) suspend Customer's access to all or part of the Services; and
(e) terminate the Agreement or the affected Order Form in accordance with the termination provisions of the Master Subscription Agreement (Section 11).
11.3 Notice. Suspension under section 11.2(d) is governed by the Master Subscription Agreement (Section 18): Taint Labs will limit it to the extent and duration reasonably necessary, give Customer prior notice where practicable and otherwise prompt notice after the fact, and restore access once the violation is resolved. Where the violation does not pose a material risk to the security, integrity or availability of the Services or to any third party, is not an Evaluation conducted without authorization or outside the Evaluation Scope, and immediate action is not required by law (MSA Section 18.1(a), (b) and (d)), Taint Labs will also give Customer a reasonable opportunity to cure before suspending. Fees continue to accrue during a suspension caused by Customer (MSA Section 18.2). [●]
11.4 Law enforcement and affected parties. Taint Labs may report conduct it reasonably believes is unlawful to law enforcement, and will report child sexual abuse material as required by law. Taint Labs may cooperate with law enforcement, regulators and, where Customer's conduct has affected them, the owners of affected systems, and may preserve relevant records for that purpose. Taint Labs will disclose Customer Data to a third party only as permitted by the Agreement and the Data Processing Addendum, or as required by law.
11.5 No duty to monitor. Taint Labs has no obligation to monitor Customer's use of the Services for compliance with this AUP, and its choice not to act on a violation is not a waiver.
12Changes to this policy
12.1 Updates. Taint Labs may update this AUP to address new attack techniques, new legal requirements, new Third-Party Service policies or new features. The version date at the top identifies each version.
12.2 Notice. Taint Labs will give Customer at least [●] days' prior written notice of a change that materially restricts Customer's permitted use of the Services, except where a shorter period is required by law or by a Third-Party Service's terms, or is needed to address a security risk. Other changes take effect on publication.
12.3 No retroactive effect. A change applies to conduct after it takes effect. It does not make a past Evaluation a violation if that Evaluation complied with the version in force when it ran.