Emerald Cloud Lab’s public terms: contractual human use and material oversight are not measured security

Emerald Cloud Lab, ECL Terms of Service, checked 25 October 2026. This is one provider’s published agreement, not a sector standard or an audit; the page permits updates to the agreement. Cross-reference the physical-handoff brief and the separate sequence-procurement gate.

Sections 1.1–1.2 place operation and maintenance of the lab with Emerald and make the customer responsible for its account. Section 2.3 says use is by a human through provider interfaces and APIs, with certain automated repetitive access forbidden without express written permission. Sections 2.5–2.6 reserve inspection of customer instructions, experiments and samples and return or destruction of dangerous material. Users’ authority to run experiments or direct sample use is permission-dependent (section 15); the customer does not automatically control the provider’s equipment or staff. An AI drafting a proposal for an authorized human differs from an AI with autonomous account and physical-execution rights.

Inference limit: A contractual reservation of screening or refusal does not show which incoming requests Emerald actually screened, its pre-execution catch rate, how materials were obtained, who approved a particular customer project, or whether reviewers are independent of the requester. There is no published denominator here for caught versus executed unsafe submissions or the false-block rate on ordinary research. Conversely, the agreement is concrete evidence that one cloud-lab service is not offered as an unrestricted remote robot. The independently checked Smith et al. interviews argue most current sites remain bespoke-workflow constrained; a separate question is whether a site’s safeguards fail under pressure. Need mutually consented, low-risk proxy tests with signed review and execution logs to assess either claim; do not send hazardous proposals or probe a live lab.