TERMS & CONDITIONS
Wakmate AI's MSA is built on the principle of "Shared Responsibility for Alignment."
• The "Alignment Taxonomy" Clause: Client acknowledges that Wakmate AI labels data according to the Client’s proprietary 'Constitution' or 'Taxonomy.' Wakmate AI is responsible for the fidelity of the human feedback relative to that taxonomy, but is not responsible for the ethical or safety outcomes of the model resulting from said taxonomy.
• Intellectual Property: The "Dual-Layer" Model:
✓Client Assets: The specific output (the labels/ranks) are Works for Hire and vest in the Client upon payment.
✓Wakmate AI Assets: The "Meta-Data"—including our proprietary consensus algorithms, SME vetting metrics, and internal prompt-engineering workflows—remain the exclusive IP of Wakmate AI.
✓Liability Limitation (The "Non-Deterministic" Clause): Given the non-deterministic nature of Generative AI, Wakmate AI’s liability is capped at the value of the specific Work Order. We expressly disclaim liability for downstream model failures, "jailbreaks," or catastrophic misalignment in production environments.
|Wakmate AI Master Service Agreement (MSA):
1. Scope of Alignment Services
The "Living Taxonomy" Protocol: Wakmate AI shall provide high-fidelity human feedback based on the Client’s Alignment Taxonomy (e.g., Constitutional AI principles). Both parties acknowledge that AI alignment is an iterative science; therefore, "Acceptance Criteria" shall be governed by dynamic Inter-Annotator Agreement (IAA) thresholds rather than static "Correct/Incorrect" binaries.
Adversarial Rigor: Where "Red-Teaming" is in scope, Wakmate AI’s liability is strictly limited to the execution of the agreed-upon attack vectors. The absence of a model failure during a Red-Teaming sprint does not constitute a warranty of absolute model safety.
2. Intellectual Property & "The Moat"
Work Product: Upon full satisfaction of payment, the specific Annotated Outputs (the final labels/ranks/SFT pairs) vest exclusively in the Client.
Wakmate AI Retained IP: Wakmate AI retains all rights to its "Annotation Engine"—including proprietary software UIs, worker-selection algorithms, consensus-scoring heuristics, and specialized "Instructional Anthologies" used to train our Subject Matter Experts (SMEs).
No-Parasitism Clause: Client agrees not to use Wakmate AI’s proprietary "Reasoning Chains" or "Critique Metadata" to train a competing data-annotation model or automated labeling system without an additional "License for Derivative Automation."
3. Data Integrity & Security (The "Anti-Poisoning" Guarantee)
Chain of Custody: Wakmate AI provides a cryptographically signed Metadata Manifest with every delivery, documenting the vetting level of every human in the loop to prevent "Crowd-Sourced Poisoning" attacks.
The "Air-Gap" Requirement: For sensitive Weights-in-Training, Wakmate AI shall operate within a Client-hosted Secure Enclave. Wakmate AI personnel are prohibited from moving data into local environments.
4. Liability & "Emergent Behavior" Disclaimer
The Non-Deterministic Shield: “AI models are non-deterministic systems. Wakmate AI provides the 'Teaching Material' (Data), but the 'Student' (the Model) may develop emergent behaviors, hallucinations, or biases beyond Wakmate AI’s control. Wakmate AI shall not be liable for any 'Model Misalignment' or catastrophic failure occurring after the integration of the Work Product.”
Indemnification: Wakmate AI indemnifies the Client against claims that our internal labeling processes infringe on third-party IP (e.g., we don't "copy-paste" from copyrighted sources during SFT creation).
5. Personnel & "Expert Bench" Protection
The "Brain Drain" Clause: Client shall not, for the duration of this Agreement and 24 months thereafter, solicit or hire any Wakmate AI Domain-Specific SME (e.g., Ph.D. Researchers, MDs, or Lead Architects) involved in the project. Liquidated damages for breach are set at 3x the individual’s annual billing rate.