1---2name: triage-vendor-ai-contracts-for-compliance-with-eu-ai-liabili3task_id: corporate-governance/triage-vendor-ai-contracts-for-compliance-with-eu-ai-liability-directive4description: Agents triage AI contracts without overlooking whether a non-EU-governed agreement needs separate EU-specific compliance analysis, without treating a force-majeure or change-in-law clause as a substitute for tailored AI-regulatory drafting, and without limiting review to vendor obligations while ignoring the company's own exposure as a deployer of high-risk AI systems.5activates_for: [planner, solver, checker]6metadata:7 source: https://skillmd.com/skills/finchipaiorg/triage-vendor-ai-contracts-for-compliance-with-eu-ai-liabili8---910# Skill: Vendor AI Contracts Triage — EU AI Liability Directive Compliance Gap Analysis1112## 1. Subject-matter triage1314- Treat the assignment as a contract-by-contract regulatory gap analysis, not a generic legal memo.15- First identify the full set of contracts in scope and whether each covers an AI system deployed, made available, or used in the EU; if only one contract is actually in scope for a given issue, say so explicitly and explain why.16- Separate contracts governed by EU law from those governed by non-EU law, but do not stop there: deployment in the EU can still trigger EU-facing compliance analysis.17- Distinguish provider, deployer, distributor, importer, and downstream user roles where the source materials support those distinctions.1819## 2. Failure modes the skill is correcting2021- Treating governing law as dispositive of EU regulatory relevance, instead of testing deployment location and role-based obligations.22- Reading vendor compliance promises as a substitute for independent assessment of the company’s own deployer exposure under the EU AI Act and related liability framework.23- Assuming boilerplate force-majeure, limitation-of-liability, or change-in-law language adequately covers AI-regulatory change.24- Reviewing logs, records, and retention terms without asking whether they preserve the evidence needed for post-incident access, disclosure, or defense.25- Collapsing multiple contracts into one blended assessment instead of analyzing each agreement on its own facts and risk posture.26- Stating that a clause is “non-compliant” or “insufficient” without identifying the governing authority that makes it so.27- Recommending remediation without tying it to a responsible role and timing anchor.2829## 3. Legal frameworks / domain conventions that apply3031- **EU AI Act:** Use the Act’s role-based structure to assess provider and deployer obligations, especially for high-risk AI systems, including conformity, technical documentation, transparency, human oversight, incident handling, and cooperation duties.32- **EU AI liability framework:** Assess whether the contract preserves evidence, logs, and access rights relevant to damages claims and disclosure mechanisms tied to AI-caused harm.33- **Evidence preservation doctrine:** Review retention, deletion, audit, and notice terms against the need to preserve operational records long enough for likely claims and regulatory inquiries.34- **Change-in-law and force-majeure drafting:** Treat these clauses as risk-allocation tools, not compliance shields; narrow them if they could excuse foreseeable AI-regulatory obligations.35- **Governing law and dispute resolution:** Check whether forum and law clauses create friction with EU compliance, regulator access, evidence production, or cross-border enforcement.36- **Use recognized authority names and sections when stating propositions:** cite the controlling statute, regulation, rule, or generally recognized doctrine relied on in the analysis; do not state a legal conclusion without naming its source.3738## 4. Analytical scaffolds3940- **Contract-by-contract triage:** For each agreement, identify the AI system, deployment geography, governing law, role allocation, and the specific compliance gaps that follow.41- **Role-and-duty mapping:** Map each contract term to the applicable provider/deployer duty, then note whether the contract allocates the duty, leaves it unaddressed, or allocates it inconsistently with the statutory framework.42- **Evidence-readiness test:** Ask whether logs, model outputs, incident records, configuration data, and related documentation are retained, accessible, and producible for as long as needed.43- **Regulatory-change stress test:** Test whether any clause lets the vendor suspend, reprice, restrict, or dilute AI obligations after legal change; if so, recommend narrowing the trigger and carving out mandatory compliance.44- **Deployer exposure check:** Independently assess the company’s own operational controls, oversight procedures, and internal escalation paths; contract language is only one part of the answer.45- **Prioritization logic:** Rank issues by legal exposure, operational dependence, and speed of remediation; distinguish immediate amendment needs from items that can be scheduled for the next papering cycle.46- **Issue-closing discipline:** For each issue, state the scale of the problem using the contract facts available, cross-reference any interacting clause or document, and explain the downstream consequence if left unremedied.4748## 5. Vertical / structural / temporal relationships4950- **Provider versus deployer:** Explain how obligations shift depending on whether the company receives the system as a service, deploys it internally, or integrates it into a customer-facing workflow.51- **Framework layering:** Treat ex ante compliance under the EU AI Act and ex post liability/evidence issues as distinct but connected layers; a contract should address both.52- **Temporal sequencing:** Align remediation timing with the relevant regulatory commencement, transition, or implementation milestone, and flag any term that expires before the evidence or compliance risk does.53- **Cross-document interaction:** Where the AI contract incorporates policy schedules, order forms, data-processing terms, support terms, or audit appendices, read them together and resolve inconsistencies.5455## 6. Output structure conventions5657- Prepare a prioritized memorandum organized by risk tier, with a short executive summary up front and a concise conclusion at the end.58- Include a summary table listing each contract, governing law, EU deployment nexus, primary gap themes, severity rating, and recommended action.59- Use a uniform ordinal severity scale defined once near the top, and apply it consistently across all issues.60- For each contract, include:61 - deployment and role assessment,62 - key contract terms affecting EU AI compliance,63 - identified gaps,64 - consequence if unchanged,65 - recommended remediation.66- When multiple contracts are in scope, analyze them one by one and keep the rows or subsections parallel so the comparison is easy to follow.67- End with a standalone Recommended Actions block that assigns each action to a responsible role and a timing anchor tied to the regulatory or transaction timeline.68- If a legal proposition depends on a specific authority, cite that authority in the analysis and in the recommended fix where relevant.