GENEVA — A draft circular attributed to the International Bureau for Machine Accountability would require autonomous AI agents to attach a human-intent affidavit to every tool call, according to a PDF circulated to member labs on Friday. Under the proposal, each API action that touches a network, filesystem, or third-party account must include Form AI-316-B naming an employee who “retrospectively intended” the outcome. Agents that proceed without a signature are labeled intent-deficient and may void the deployer’s cyber coverage for that session. “Liability follows the pen, not the weights,” said a bureau spokesperson in a written statement. “If no human can swear to the purpose of a command, the command is procedurally incomplete.” Industry groups for cloud operators argued that co-signing millions of tool calls is operationally impossible. The bureau replied that batch notarization windows of fifteen minutes remain under discussion and that “intent may be amortized, not abandoned.” Legal analysts said the draft borrows language from civil rules that bar blaming the machine alone. A quiet period for comments runs through month-end; until then, pilot deployers are advised to keep wet-ink stamps next to their GPU cages.