The Race — Building Democracy in the Age of AI. People race toward a shared finish line.
The Race

Policy tool · See record

Agent action logs and delegated-authority controls

What the tool does

Bucket: Accountability and Due-Process Tools / Anti-Lock-In Tools Status: Public tool / active tool candidate Primary lanes: Lane 05 — Platforms, Distribution, and Separations; Lane 01 — Governance and Regulation

Current status

An existing instrument family, updated with the approved example below; it does not establish independent public oversight by itself.

Where it has been used

GSA’s AI purchasing protections

Added by

GSA adopted a standard contract clause on September 28 for government purchases in which a large language model is a material feature and directly receives or produces government data. Mandatory use in new covered purchases starts October 19; contracting officers may use it earlier. Incorporating it into an existing contract is a separate contracting decision. This is an adopted purchasing rule, not a single new supplier contract that every government can immediately order from.

The clause protects covered government inputs, outputs and custom developments; limits their reuse for training, sale and marketing; requires portable data and documented interfaces; and protects migration from obstructive charges. It includes reasonable human intervention and action traces, incident reporting within 72 hours, ordinarily 30 days' notice of material changes, and suspension and noncompliance remedies. Supplier ownership of pre-existing models and technology remains. Decommissioning liability after qualifying termination is capped at 25% of the affected order; replacement-system and reprocurement costs are excluded.

Eligible state, local and tribal governments and public educational institutions can use relevant GSA purchasing programmes within their statutory scope. If the underlying eligible contract contains this clause and the supplier accepts the non-federal order, its applicable terms carry into that separate purchase. Suppliers may reject those orders; eligibility, product category and incorporation must be checked for the particular purchase. The protection is therefore potentially reusable well beyond federal agencies, without assuming universal coverage or automatic amendment of all existing contracts.

Read more about this case →

AI evaluations enter a procurement court record

Added on

In the TRAX procurement challenge, the Court of Federal Claims required three AI-generated proposal evaluations to be included in the record on July 10. Its later merits opinion, made public September 22, described their circulation to people involved in the Army's evaluation of a roughly $449 million support contract. An October 6 legal explanation brought the case back into view.

TRAX ultimately lost its challenge. The court's treatment of the documents did not establish that AI evaluation is always lawful or unlawful; it showed that the government's description of how material was used can be tested against the actual record. The distinction between generating advice, sharing it and relying on it mattered.

Read more about this case →

Microsoft’s agent permission containers

Added on

Microsoft announced general availability of Microsoft Execution Containers on October 7. Developers and administrators specify which files, network connections and other resources a workload can use. The container enforces those limits outside the agent’s own control, so generated code cannot simply grant itself more access.

For example, a coding agent can be allowed to edit its repository while being blocked from personal documents or network destinations. The common configuration maps to different containment backends on Windows, macOS and Linux. Some options are platform-specific, and the hardware-isolated microVM option is experimental. Announced Entra identity, Intune management and expanded Agent 365 controls include future features; they should not be described as generally available today. Microsoft technical account.

Read more about this case →

Personal Agent Protocol

Added on

Personal Agent Protocol's public documentation now identifies draft version 0.1, updated October 9. Sierra announced the initiative on October 6 with Meta and other participating businesses. The draft describes how a user's AI assistant could identify itself to a business and ask to perform permitted tasks.

A business publishes a standard discovery file explaining its interfaces. A person can then authorize an agent with limited permissions through an authorization process, rather than handing over a full account password. The specification supports interaction through websites, APIs or business agents, with a route to human help. These are draft specifications: broad deployment, reliable cross-provider operation and actual purchasing permissions have not been demonstrated by their publication. announcement and participants. Draft permissions/interoperability, not demonstrated deployment or independence.

Read more about this case →

Proof’s agent identity wallet

Added on

Proof released x401 on October 15, 2026 as an identity wallet, open web protocol and assistant connection. A merchant can request signed evidence of identity and transaction terms, then verify whether the instruction still matches the person’s approval. Proof is a commercial credential issuer; promised neutral protocol governance is not already established.

Read more about this case →

GovWare’s agent-commerce demonstration

Added on

Separately, an October 14 GovWare demonstration, described in an October 15 technical account, linked discovery, identity, authorization, communications, payments and audit across platforms. It used a controlled flight-purchase scenario, not evidence of general production commerce or actual settlement at scale. The technical paper describes the architecture.

Read more about this case →

Mastercard demonstrates delegated spending controls

Added on

Mastercard’s October 20 Agent Pay for Machines demonstration and early-access interfaces let a business assign an agent a task and spending limit, with signed vouchers and seller verification. This is a pilot and early access, not generally available or issuer-neutral production commerce. Admission, revocation, logs and production terms remain important conditions.

Read more about this case →

Stripe adds bounded agent authority to move funds

Added on

Stripe announced public-preview write access for financial agents on November 19, 2026. An eligible customer can authorize a client through OAuth, where each money-moving action requires human confirmation, or use an agent API key with its own identity, permissions, expiry and rotation. Agent keys default to approval for refunds, payouts and transfers, but customers can customize rules, such as permitting ACH transfers below $50 without individual approval.

Access depends on Stripe Treasury eligibility, the account and the client implementation. These controls are distinct from the previously recorded October read-only Treasury access. Preview documentation does not establish independent performance, portable audit records or interchangeable banking backends.

Read more about this case →
Additional source perspective: Agent action logs and delegated-authority controls

Status: Public tool / active tool candidate Current anchors: FIDO, NIST, Mastercard, Arcade.

Use as a mature public-facing tool cluster.


Source snapshot: 2026-11-20. Individual event and verification dates remain as recorded above. Inclusion alone does not assign a score.