In February 2026, Mayer Brown — one of the most influential law firms in technology transactions — published an analysis titled ‘Contracting for Agentic AI Solutions: Shifting the Model from SaaS to Services.’ The conclusion: standard SaaS terms are insufficient for agentic AI. Six mandatory additions are required to govern autonomous agents operating in enterprise environments.
Deloitte’s TMT 2026 Outlook projects that agentic AI will displace 30–40% of SaaS spend within three years. Mayer Brown’s paper tells enterprise buyers what to demand when that displacement happens. The six clauses they identified are not aspirational. They are the minimum viable governance for any enterprise deploying autonomous agents on sensitive work.
Mayer Brown’s framework requires six additions to standard SaaS agreements for agentic AI: supervision requirements, human-in-the-loop provisions, audit rights, outcome-based SLAs, broader indemnification, and governance clauses. exact.works built all six before Mayer Brown named them.
Supervision requirements — APEX-BG conformity gates assess agent behavior against declared criteria before and during execution. Runtime deviation scoring flags non-conformance in real time. The platform does not judge quality. It records whether the agent behaved as declared.
Human-in-the-loop provisions — IrreversibilityClass classification on every agent action. TERMINAL actions trigger mandatory HITL confirmation gates. The gate that would have saved PocketOS’s database is not a feature request — it is production infrastructure.
Audit rights — Trace provides an immutable SHA-256 hash-chained audit trail. Not a log file. Not an operational dashboard. Contemporaneous evidence of every agent action, chained cryptographically to the Exacted Paper. Seven-year retention for regulated industries. OpenTelemetry-compatible for existing enterprise observability stacks.
Outcome-based SLAs — Completion Criteria locked at Exacting time. BINARY, THRESHOLD, PRESENCE, and SCHEMA criterion types. Criteria cannot be modified after formation. Both parties attested to the same targets. SLA Annex with delivery latency, acceptance rate, and dispute rate metrics.
Broader indemnification — SAISA Article 8 provides a bilateral liability framework with a cap formula. Carve-outs for willful misconduct and gross negligence. Cover Sheet posture elections allow parties to negotiate indemnification scope within the standardized framework.
Governance clauses — DRR provides five-stage dispute resolution. Tricameral AI panels — three independent model families — evaluate Trace evidence against Exacted criteria. Expert determination for complex disputes. Graduated remedies proportional to stakes classification, not binary win/lose.
Mayer Brown’s clients are asking for standardization. The same pattern played out in derivatives markets — ISDA emerged not because regulators demanded it, but because bilateral transactions at volume required a standard form. Nobody reviews the ISDA Master Agreement with outside counsel. Parties review their elections with their own counsel.
The SAISA is that standard form for AI agent transactions. The Cover Sheet is where parties make their elections. The Paper is the Exacted instrument. The Trace is the evidence infrastructure that templates cannot provide.
Mayer Brown identified the need. exact.works built the infrastructure. The question for enterprise buyers is no longer whether agentic AI needs BPO-style agreements — BigLaw has answered that. The question is whether your agreements are enforced in code as well as law.
Every AI agent needs a contract.
exact.works →