Hybrid Contracts Advised for Agentic AI
Legal experts are advocating for new hybrid contract models for deploying agentic AI solutions, moving beyond standard SaaS terms. An analysis from law firm Mayer Brown suggests blending software usage terms with service-level objectives, outcome-based guarantees, and explicit risk-sharing clauses. This approach is considered particularly crucial for public sector and political technology, where accountability and auditability are paramount.
- Standard Software-as-a-Service (SaaS) agreements are ill-suited for agentic AI because they define the "service" as access to a platform, making the user responsible for all actions; a hybrid model redefines the service as the tasks the provider completes using AI agents. - The EU's AI Act, which began phased implementation in 2024, classifies AI by risk level and imposes strict obligations on "high-risk" systems, a category many agentic AI solutions in finance and public services fall into. This necessitates contractual clauses for transparency, auditability, and human oversight to ensure compliance. - A key shift in hybrid contracts is moving from technical uptime guarantees (like 99.99% availability) to operational, outcome-based Service Level Agreements (SLAs) that measure performance on metrics like accuracy. For example, an SLA might require 99% of invoices to be processed correctly against a purchase order. - Traditional SaaS contracts often limit the provider's liability to intellectual property infringement claims. In contrast, hybrid models inspired by Business Process Outsourcing (BPO) include broader indemnities, holding the provider accountable for third-party claims arising from the agent's autonomous actions, such as discrimination in an automated hiring process. - Liability allocation is a central challenge, as it can be difficult to trace an error back to a single source in a complex AI system. Hybrid contracts explicitly define responsibility for AI errors, data privacy, and compliance, often requiring specific AI liability insurance coverage. - The "black box" nature of some AI makes transparency and audit rights critical contractual elements. These rights allow the customer to review the AI's processes, data, and outputs to mitigate risk and ensure the system's decisions can be explained to regulators and end-users. - In the UK public sector, where annual procurement spending exceeds £380 billion, there's a significant push to link payments for AI systems to tangible outcomes like cost savings or improved service delivery, rather than just licensing fees or API calls. This aligns the financial incentives of the AI provider with public value. - Agentic AI can cause "model drift," where the system's accuracy and behavior change over time as underlying data shifts. Contracts must account for this by including provisions for continuous monitoring, auditing, and performance management to ensure the AI remains aligned with its original goals.
