Summary

  • In May 2014, ICANN accepted Governmental Advisory Committee consensus advice and changed three .AMAZON applications into a practical stop. The advice carried a “strong presumption” that the Board should not approve the applications, but the Applicant Guidebook did not make that presumption conclusive and did not transfer the Board’s decision duty to governments.
  • Amazon’s 2014 reconsideration request did not operate as a merits appeal: the then-applicable standard asked whether material information had been ignored or materially false or inaccurate information relied upon. The majority in the 2017 Independent Review Process reached the institutional question that reconsideration could not: a Board may accord substantial weight to consensus advice, but it must still exercise its own reasoned judgment and give reasons capable of review.
  • The IRP did not award the strings, order a registry agreement or direct a root-zone change. It declared the earlier decision method inconsistent and recommended a renewed exercise of judgment. ICANN then sought further policy information, facilitated negotiations, removed the formal “Will Not Proceed” status, accepted a package of Public Interest Commitments and only later completed contracting and delegation.
  • The eventual registry contract, whose current operator is Amazon Registry Services, Inc., converted broad claims about regional identity and public interest into narrower contractual rights: reserved terms, nine non-commercial names for ACTO and its member states, and specified standing to invoke a contractual dispute procedure. Those rights were enforceable, but they did not amount to co-control of the registry, and their acceptance does not prove that every governmental or regional objection was resolved.

The status change that was not a final legal answer

On 14 May 2014, three applications stopped moving. The New gTLD Program Committee of the ICANN Board accepted consensus advice from the Governmental Advisory Committee and directed that the applications for .amazon, .亚马逊 and .アマゾン “should not proceed”. In ICANN’s application machinery, the consequence was expressed as a status: Will Not Proceed. The Board resolution did not cancel the 2012 programme, amend the root or transfer a registry. It did something more immediate. It instructed the organisation that controlled processing to stop carrying the applications towards contract.

That distinction matters because the dispute was never governed by a single sovereign command. Brazil and Peru could issue an early warning. Governments acting through the GAC could reach consensus advice. The ICANN Board could accept or reject that advice and direct application status. ICANN organisation could process or pause the files. An IRP panel could declare whether ICANN had acted consistently with its governing commitments and recommend a response. Contracting staff could execute registry agreements only after the application path reopened.

The IANA functions could process delegation only after the approved and contracted party satisfied the root-zone requirements. No one institution held every lever.

The May 2014 decision nevertheless had the practical force of a veto because the institution that did hold the next lever—the Board—treated the governmental consensus as sufficient to end processing. Three years later, the Amazon IRP Final Declaration found, by a two-to-one majority, that ICANN had converted a strong presumption into complete deference. The panel majority declared the prior decision inconsistent and recommended that the Board return to the question, make an objective and independent judgment about whether well-founded, merits-based public-policy reasons justified denial, and explain any renewed refusal. That declaration reopened decision, not ownership. It neither granted the applications nor instructed IANA to place the strings in the root.

The five-year hold in the title therefore needs precision. The formal Will Not Proceed status lasted from May 2014 until the Board removed it in October 2018. But removal of the status did not produce an immediate contract. Facilitation and proposal exchanges continued, the Board did not accept Amazon’s revised commitments proposal until May 2019, and the registry agreements were signed in December 2019. Consensus advice held the applications in a broader institutional sense: it shifted the burden, changed the sequence and kept access to the contracting gate contingent on a political-policy dispute that ordinary technical evaluation had not resolved.

This was not simply a clash between a company and countries over a word. It was a test of how governmental participation in ICANN becomes operational control. Governments did not need a treaty conferring ownership of the string. They needed a recognised procedural instrument, a consensus rule and a Board willing to let that instrument determine application status. Amazon did not need an IRP panel with power to award the namespace. It needed a declaration that the Board’s own constitutional duties survived the presumption. Recovery then depended on the institution that had stopped the applications in the first place.

Three applications entered several different decision tracks

Amazon EU S.à r.l. applied in 2012 for the ASCII string and two Internationalised Domain Name versions. The principal .amazon application was not rejected by the programme’s ordinary technical and financial tests. Its Initial Evaluation report, dated 12 July 2013, recorded passes for background screening, string similarity, DNS stability, registry services, technical and operational capability, and financial capability. The Geographic Names Panel also concluded that the applied-for label was not a geographic name under the definitions in the 2012 Applicant Guidebook.

That pass answered a bounded question: whether the application met the programme criteria assigned to Initial Evaluation. It did not decide whether governments could raise a public-policy objection through the GAC. Nor did the finding that .amazon was not a protected geographic name establish that geographic, cultural or political interests were irrelevant. The programme had parallel routes. A string could pass technical evaluation, survive a formal objection and still encounter GAC advice addressed directly to the Board.

The formal objection track illustrates the separation. In March 2013, the Independent Objector filed community objections against the three applications. In January 2014, the objection panel rejected them, concluding that the required showing of substantial opposition or substantial detriment had not been made. The May 2019 Board record later recited that Amazon had prevailed in each objection. That result was important evidence, but it did not bind the GAC or extinguish the Board’s authority to consider advice under a different rule.

The same is true of the legal analysis commissioned during the 2014 decision process. According to the 14 May 2014 rationale, Professor Jérôme Passa concluded that no applicable international, regional or national legal rule obliged ICANN to reject the applications, and none obliged it to accept them. The analysis removed the claim of legal compulsion in either direction. It did not eliminate discretion. Once law did not dictate the result, the quality of the institution’s reasoning became more, not less, important.

This is the first institutional lesson of the case. Formal programme success is not control over the whole process. Amazon could participate in evaluation, answer objections and present legal argument. Those acts generated evidence and a procedural position, but they did not control the Board resolution that determined whether processing continued. Conversely, governments could shape the outcome without possessing the registry contract. Their power came from the weight assigned to a recognised form of advice.

An early warning was notice, not a veto

The governmental path began with a document whose limits were stated on its face. On 20 November 2012, Brazil and Peru issued a GAC Early Warning concerning the .amazon application, with support recorded from Bolivia, Ecuador, Guyana and Argentina. They argued that exclusive private rights in the label would prevent public-interest use associated with protecting, promoting and increasing awareness of the Amazon biome; could hinder the aggregation of web pages connected with the region and its population; and implicated the English name of the Amazon Cooperation Treaty Organization.

Those were governmental claims, not findings of fact adopted by a tribunal. The early warning did not itself reject the application. The GAC process described it as notice that one or more governments had identified a public-policy concern and that later GAC advice or a formal objection was more likely. The warning gave Amazon information about the risk and an opportunity to withdraw or respond. It changed incentives before it changed legal or operational status.

That difference between notice and decision is easy to lose because the same governments later helped form the consensus that carried the strong presumption. Yet the early warning’s design was deliberately non-dispositive. It allowed a government to place a concern into the record without proving the elements of a formal objection and without securing agreement from the full GAC. Its authority lay in escalation. A warning signalled that the applicant’s technical pass would not be the only relevant file.

Amazon responded with legal and policy arguments and offered commitments. The 2014 Board rationale records proposals to reserve culturally sensitive terms for ACTO and its member governments, continue dialogue and refrain from objecting to future applications for related labels such as “Amazonia”, “Amazonas” or “Amazonica”. These offers show that bargaining began before the Board’s stop decision. They also show why participation cannot be equated with remedy: an applicant can answer every concern and still lack a mechanism that compels the decision-maker to explain why the answer failed.

For the governments, the early warning created leverage without final responsibility. They could define the public-policy concern, invoke geography and regional identity, and make withdrawal the requested cure. But they did not have to issue a reviewable administrative decision under a national statute. The Board remained the institution whose action changed the application. That allocation made the quality of the Board’s adoption of governmental advice central to the later IRP.

Consensus converted concern into a presumption

The 2012 Applicant Guidebook created a special legal-procedural effect for GAC consensus. Module 3 separated GAC advice from the programme’s formal objection proceedings. It contemplated three forms. Consensus advice that an application should not proceed created a strong presumption that the Board should not approve it. Advice identifying concerns required dialogue and reasons. Advice that an application should not proceed unless remediated created a strong presumption unless an available cure was implemented. The applicant received 21 calendar days to respond after publication of the advice.

The rule was powerful precisely because it stopped short of saying “automatic rejection”. A presumption assigns a starting position and a burden. It can require unusually persuasive reasons to depart. It does not, without more, erase the decision-maker’s duty to examine whether the premise fits the case. The Guidebook’s wording preserved the Board as the approving authority and elsewhere recognised its ultimate responsibility for the programme. Consensus increased the weight of governmental input; it did not formally substitute the GAC for the Board.

In July 2013, at ICANN’s Durban meeting, the GAC reached consensus that the three applications should not proceed. That did not make ACTO or any government a legal co-owner of the strings, and the record does not show affirmative endorsement of every policy assertion by every government. The United States had indicated that it would remain neutral rather than entity to consensus, a point later recounted in the 2014 reconsideration record. Neutrality could therefore be institutionally decisive: it allowed the advice to qualify for the Guidebook’s strongest presumption without amounting to an affirmative vote for each stated objection.

The GAC did not supply a detailed merits rationale with the Durban consensus. In March 2014, it pressed the Board to settle the matter under the Guidebook’s high-priority treatment of GAC advice. The absence of reasons did not make the advice invalid under the rules then in force. It did, however, place a heavier explanatory burden on the Board. If the advisory body did not show its reasoning, the deciding body had to reveal its own route from the record to the result.

Consensus therefore had two different effects. Formally, it triggered a strong presumption. Practically, it assembled governmental positions into a single signal that the Board was politically and procedurally reluctant to reject. The first effect was written in the Guidebook. The second arose from incentives: disagreement with consensus advice risked a visible confrontation with governments, while acceptance could be described as adherence to the programme’s governmental safeguard. The IRP later examined what happened when the practical incentive overwhelmed the formal reservation of independent judgment.

The Board accepted the advice without showing the decisive balance

The 14 May 2014 resolution did more work than a bare endorsement. The Board committee acknowledged that the GAC had supplied no rationale, drew on the Brazil-Peru early warning, reviewed Amazon’s responses and proposed commitments, considered the failed community objections and obtained the independent legal analysis. It also stated that the Board retained ultimate responsibility and that each application required individual consideration. The record was not empty.

The weakness lay in the transition from evidence to result. The committee accepted the consensus advice and directed the President and chief executive that the applications should not proceed, while leaving open continued dialogue between Amazon and the relevant GAC members. It did not explain which public-policy proposition it found established, why Amazon’s rebuttal and proposed safeguards were limited public evidence, or how it weighed the contrary results from Initial Evaluation, the community-objection process and the legal analysis. A list of materials considered is transparency about inputs.

It is not necessarily accountability for the conclusion.

That distinction is especially important where a rule creates a presumption. A reasoned decision should identify the presumed proposition, the evidence supporting it, the evidence offered in rebuttal and the standard for deciding whether the burden has been met. The 2014 resolution instead allowed the strong presumption to do most of the visible analytical work. The Board said it had considered the record and accepted the advice, but an affected applicant could not reconstruct the operative test from the published rationale.

The “without prejudice” reference to continued dialogue did not cure the deficiency. Dialogue is participation. It gives parties a place to make offers, state concerns and seek compromise. It does not confer a right to a decision, a deadline, a reasoned response or a remedy if one side refuses. After the status became Will Not Proceed, Amazon’s bargaining position was structurally weaker: the company needed an affirmative institutional act to recover processing, while the objecting governments could preserve the existing outcome by withholding agreement.

The Board’s decision also demonstrates why the control chain must be reconstructed in stages. The GAC’s consensus did not itself alter the application database. The Board resolution authorised the operational consequence. ICANN organisation implemented it. The contract gate remained closed. No root-zone actor needed to reject a delegation request because the applications never reached delegation readiness. Upstream procedural control prevented any downstream technical question from arising.

The policy concerns were not frivolous merely because the reasoning was insufficiently exposed. ACTO member states were asserting interests in a region, a treaty organisation’s name and the ability of public institutions and communities to use an internationally recognised term. Amazon was asserting the integrity of a published application programme, the legal distinctiveness of its trademark and a right to have rebuttal evidence assessed. The institutional defect was not that the Board chose one side. It was that the published decision did not make clear how the Board, rather than the existence of consensus itself, chose.

Reconsideration provided review access but not a merits remedy

Amazon filed Reconsideration Request 14-27 on 30 May 2014. It sought reversal of the stop decision, rejection of the GAC advice and a direction that staff resume processing. The request argued, among other things, that the advice arrived too late, that the Board had treated a rebuttable presumption as decisive and that the result conflicted with applicable law and programme rules. The Board Governance Committee recommendation, adopted by the New gTLD Program Committee on 8 September 2014, rejected the request.

The denial did not establish that every underlying argument lacked merit. It reflected the design of the reconsideration mechanism then available. The committee framed its task as asking whether the Board had acted without considering material information or had relied on materially false or inaccurate information. It was not a general appeal in which a reviewer could reweigh the policy record and substitute a better conclusion. Because the 2014 resolution identified the early warning, Amazon’s submissions, the legal opinion and other materials, the committee found no qualifying failure under that narrow gateway.

Timing arguments met a similar boundary. Amazon contended that GAC advice should not have arrived after the formal objection period. The committee answered that the GAC’s role arose from ICANN’s Bylaws and the Guidebook’s separate advice mechanism, not solely from the objection calendar. The advice therefore had to be considered even though the ordinary objection window had closed. This preserved governmental access, but it also confirmed that an applicant could clear one deadline only to encounter a different institutional route without the same evidentiary elements.

Reconsideration thus supplied process without effective relief for the central complaint. Amazon was eligible to file, received a written recommendation and obtained a Board-level disposition. Those features made the mechanism visible and reviewable in a formal sense. Yet the available standard did not reach the alleged conversion of presumption into deference unless that conversion could be restated as omitted or false information. Review access existed; an enforceable merits remedy did not.

The case is a useful warning against treating the number of accountability mechanisms as a measure of accountability. A reconsideration process can be correctly administered and still be incapable of testing the most important institutional wrong. The relevant question is not merely whether an affected party can submit a request. It is whether the reviewer has authority to identify the contested duty, apply an appropriate standard and order or induce a consequence that changes the decision chain.

For Amazon, the reconsideration denial left the application status intact. It did not compel a new balancing exercise, did not require the GAC to provide reasons and did not reopen contracting. The company’s next consequential route was the Independent Review Process, whose function was not to decide who deserved the string but to test whether ICANN’s action conformed to its Articles, Bylaws and established policies.

Independent review reached the duty that reconsideration could not

Amazon initiated the IRP in March 2016. The dispute was heard by a three-member panel chaired by Robert C. Bonner, with Robert C. O’Brien and A. Howard Matz. After a merits hearing on 1 and 2 May 2017, the panel issued its Final Declaration on 11 July 2017. The majority found Amazon to be the prevailing party and concluded that ICANN had acted inconsistently with its governing commitments.

The majority’s central finding was institutional rather than proprietary. The Guidebook’s strong presumption did not authorise complete deference. The Board had a duty to determine whether well-founded, merits-based public-policy reasons supported denial, and to do so through objective and independent judgment. Because the GAC had provided no rationale, the Board’s own explanation carried special importance. The majority concluded that the Board had effectively allowed the consensus advice, standing alone, to supplant the reasoned decision that ICANN’s rules required from the institution holding final authority.

The declaration also addressed procedural fairness in the GAC process. The majority considered that Amazon had not been given an adequate opportunity to make a written submission directly to the GAC before the consensus advice was formed, and treated that limitation as reducing the deference the Board could reasonably accord the advice. This part of the reasoning was contested. Matz, concurring in the result and partially dissenting, disagreed with the majority’s procedural-fairness analysis and believed the Board had undertaken substantial independent inquiry.

His narrower route was that, after Amazon had rebutted the presumption, ICANN lacked a well-founded basis for denial. The split matters: the declaration did not present a unanimous theory that every feature of GAC procedure violated ICANN’s rules.

What united the outcome was the rejection of an irrebuttable reading. Even a dissenting or partially concurring view did not treat “strong” as “conclusive”. The Board could give consensus advice exceptional weight, but weight is meaningful only if a decision-maker can identify evidence strong enough to overcome it. Otherwise the word “presumption” disguises a transfer of final authority.

The remedy was carefully bounded. The panel recommended that the Board promptly re-evaluate the applications, make an objective and independent judgment about whether well-founded public-policy reasons existed, and explain any renewed denial. If the Board decided the applications should proceed, it was to meet and confer with the GAC within 60 days. The declaration did not direct ICANN to approve the applications, execute a registry agreement or delegate the strings. It identified the Board’s decision duty and recommended that the applications be reconsidered under the proper standard.

That boundary was not a technicality. An IRP panel’s institutional legitimacy depended on testing ICANN’s compliance without becoming the programme administrator. It could declare inconsistency, recommend reconsideration and allocate costs. It did not possess the application-processing systems, contracting authority or root-zone role necessary to implement the end state Amazon wanted. The Board’s September 2017 response accepted the prevailing-party and reimbursement portions of the declaration: ICANN was to bear the administrative and panel costs and reimburse Amazon US$163,045.51, while each side bore its own legal expenses. Monetary reimbursement marked prevailing-party status; it was not a substitute for the namespace.

The declaration therefore separated procedural victory from operational control. Amazon won a declaration that the way ICANN had reached the stop decision was inconsistent with its governing commitments. It did not win a command addressed to the registry database or the root. The control holder after review remained the Board, because only the Board could direct staff to resume processing and approve the path to contract. Independent review reopened the gatekeeper’s duty; it did not remove the gatekeeper.

The Board reopened the evidence file before it reopened the application path

ICANN’s first material response was not to approve the strings. On 29 October 2017, the Board asked the GAC to provide any merits-based public-policy reasons, or new and additional information, that supported the earlier advice. It set the end of ICANN61 in March 2018 as the requested deadline. This step acknowledged the IRP’s central point: the Board needed a record capable of sustaining independent judgment.

The GAC’s operative response moved away from a bare instruction that the applications should not proceed. In its November 2017 Abu Dhabi communiqué, it advised the Board to continue facilitating negotiations between Amazon and the ACTO member states with a view to a mutually acceptable solution that would allow use of .amazon. The Board accepted that advice on 4 February 2018 and directed ICANN’s chief executive to facilitate. In March 2018, the GAC said it had no additional information beyond the Abu Dhabi position, a sequence later set out in the 10 March 2019 Board resolutions.

This was an important change in the mechanism of power. The 2013 consensus had operated as a presumption against approval. The post-IRP advice made negotiation the preferred route towards approval. It did not concede Amazon’s merits case or recognise a right to delegation. It changed the default from stop to facilitate. In its October 2018 rationale, the Board expressly treated the Abu Dhabi advice as superseding the earlier Durban advice. The Board could therefore comply with the operative advice while seeking terms that translated public-policy concerns into conditions of operation.

Negotiation, however, created its own control problem. A requirement to pursue a “mutually acceptable” solution can become an indefinite hold if either side possesses an effective veto over agreement and the decision-maker does not set an endpoint. Amazon needed processing to resume. ACTO member states could argue that delegation required their consent and that they should participate in governance. The Board could facilitate but could not manufacture consent. Unless it retained authority to decide after a defined period, the decision duty identified by the IRP could dissolve into open-ended diplomacy.

The Board began to reclaim that authority in 2018. Amazon submitted updated proposals, while ACTO member states stated on 5 September 2018 that the proposal was inadequate, that delegation required the Amazon countries’ consent and that they had a right to participate in governance. Those were the states’ positions, not rights adjudicated by the Board. On 25 October 2018, the Board removed the Will Not Proceed status and directed that the applications resume processing under the 2012 programme, including publication of proposed Public Interest Commitments.

Removal of the status was a real remedy because it changed what ICANN organisation could do. It was not yet an award. The applications still faced proposal review, public comment, contracting and delegation checks. The Board also continued facilitation, showing that a formal processing restart and political negotiation could run in parallel. The key shift was that lack of a negotiated settlement no longer automatically preserved the 2014 stop.

In March 2019, after more than a year of facilitated engagement, the Board allowed a final short period for the parties to submit a joint solution or jointly request more time. No joint request or joint proposal followed. Amazon submitted a revised commitments package on 17 April. On 15 May 2019, the Board found the proposal acceptable and directed continued processing, including publication of the commitments for comment.

The May decision is where independent judgment became operational. The Board did not claim that ACTO had consented. It acknowledged that no mutually acceptable solution had been reached. It instead decided that the proposal was not inconsistent with the operative GAC advice and that no public-policy reason required the applications to remain blocked. That conclusion may be debated, but it was a Board conclusion rather than a mechanical consequence of consensus or non-consensus. The institution holding decision authority finally made the choice that the IRP had said it must make under ICANN’s governing commitments.

That decision itself triggered another accountability proceeding. Colombia filed Reconsideration Request 19-1 on 15 June 2019 and asked ICANN to halt contracting and possible delegation while its objections were examined. ICANN’s normal process held those downstream steps pending disposition of the request. On 14 August, the Board Accountability Mechanisms Committee recommended denial, concluding that the Board had acted on accurate and complete information and consistently with the Bylaws and programme procedures. The Board adopted that recommendation on 8 September, a sequence recorded in ICANN’s December status update. The request temporarily preserved the status quo, but it did not create a right to joint governance or require the Board to reverse its May judgment.

Public-interest claims became narrower contractual commitments

The 2019 package did not resolve the philosophical question of who “owns” the name Amazon. It converted parts of the dispute into obligations that a registry contract could identify and enforce. The May 2019 Board record described a proposal that included a joint steering committee as well as three principal operational commitments. Amazon would not use as domain names terms whose primary and well-recognised significance concerned the culture and heritage of the Amazonia region. It would provide nine names in each top-level domain for non-commercial use by ACTO and its eight member states. It would also permanently block up to 1,500 culturally significant terms in each registry. The executed agreement retained those name-use and reservation obligations, but its ACTO commitment contains no steering committee and assigns no vote over registry management.

These terms were materially different from the governments’ broad claim that delegation required country consent or that ACTO states should participate in governance of the top-level domain. A reservation list can be administered. A quota of names can be allocated. A dispute about whether a requested label falls within a defined class can be taken to a contractual process. By contrast, “participation in governance” is indeterminate unless the agreement specifies voting rights, approval thresholds, management authority, information rights and remedies.

The contract settled on the former kind of obligation, not the latter kind of shared control.

The registry model remained highly restricted. Amazon said it intended .amazon and the IDN equivalents to operate as brand registries, with registrations aligned to its global brand strategy rather than open retail registration. That reduced one category of governmental concern: unrelated third parties would not have a general market in which to register regional or culturally sensitive names. It also reinforced Amazon’s control, because the operator—not a multistakeholder regional body—would decide which brand-aligned names to create, subject to the commitments.

After Amazon submitted the commitments through the application change process on 5 June 2019, ICANN published them for comment. The 19 December 2019 status update says the comment period ended on 12 October, ICANN reviewed the comments and concluded that the applications could proceed to contracting. Publication made the proposal visible. It did not give commenters a vote or require staff to produce an adjudicative report answering each submission. Again, participation and control remained separate.

The final .amazon Registry Agreement placed the ACTO commitments in Specification 11. It supplied the element that earlier dialogue lacked: a legal entity against which performance could be tested. The agreement required the operator to comply with the commitments and connected serious breach to ICANN’s contractual enforcement powers. The Public Interest Commitment Dispute Resolution Procedure offered a defined route for complaints about non-compliance.

Standing was not universal. Apart from ICANN’s own contractual enforcement authority, the ACTO commitment identified ACTO and its member states as the sole entities entitled to invoke or contest an alleged failure through the PIC procedure. That is a consequential right. It allows named governmental institutions to move from political objection to a contractual claim. But it does not give every Indigenous group, regional resident, civil-society organisation or business an equivalent direct remedy. A public commitment can be transparent to all while enforceable at the instance of only specified parties.

The operating detail further limits what “shared use” means. The agreement permits ACTO and its member states to request designated names, but the operator may assess whether a request fits the agreed categories. ACTO’s authorised representative had two years from the agreement’s effective date to submit additions to the list of up to 1,500 requested-reserved names; the accepted list remains reserved for the agreement’s term.

The operator registers permitted names through its chosen registrar and places them on servers it selects and controls; content must be supplied through the agreed arrangement; use is non-commercial; and technical and content restrictions apply. ACTO can challenge an adverse determination through the specified procedure, but it does not administer the registry’s technical systems, select the backend provider, control the root-zone contacts or authorise the operator’s other registrations.

Functionally, the package created bounded access, reservation rights and enforcement standing. It did not create a joint registry. That is a reasoned inference from the allocation of contractual powers, not a claim that the commitments lacked value. The rights may protect names and provide institutional visibility for the region. They also leave the operator with the decisive operational position. The dispute moved from whether governments could stop the namespace to which constraints they could enforce once the private operator received it.

Nor does execution of the agreement prove that all affected communities accepted the bargain. The public record shows continuing ACTO objections and the absence of a mutually acceptable joint solution before the Board moved forward. Private negotiations and internal governmental deliberations are not fully public. The most that the official record establishes is that ICANN found the package sufficient for its programme and embedded it in the contract. Contractual sufficiency is not the same as political consensus.

Contract execution and root-zone delegation were separate acts

On 19 December 2019, ICANN announced that it was signing registry agreements and Specification 13 brand amendments for the three applications with Amazon EU S.à r.l. The original .amazon Registry Agreement names that company as registry operator. ICANN’s current agreement record lists Amazon Registry Services, Inc. as operator while preserving the 19 December 2019 agreement date. Contracting gave the operator enforceable rights and duties against ICANN. It did not by itself insert the label into the DNS root.

Delegation required a later root-zone process. IANA’s delegation report for .amazon, dated 2 June 2020, records that the application process was complete, the proposed manager matched the approved party, contacts were confirmed, technical conformance was completed and the remaining procedural checks had been passed. The current IANA record gives a registration date of 28 May 2020 and lists Amazon Registry Services, Inc. as sponsoring organisation. Separate IANA records establish the same registration date and sponsor for .亚马逊 and .アマゾン.

The sequence matters for accountability. A Board resolution can authorise processing but cannot substitute for a registry agreement. A registry agreement can define commitments but cannot substitute for technical and administrative delegation readiness. A root-zone record can show who sponsors a delegated top-level domain but cannot establish whether the preceding policy bargain was fair or whether its predicted social effects occurred. Each institutional entity answers a different question.

The present delegation records prove operational control in the narrow DNS sense: the sponsoring organisation is responsible for managing the delegation details, and the strings exist in the root under that sponsorship. They do not prove that the reserved-name regime has delivered the cultural or economic benefits claimed by either side. Current registration use, even if observable, would not by itself establish those effects. The governance outcome is clearer than the social outcome.

What the control chain reveals

The full chain can now be stated without collapsing its stages. Brazil and Peru used an early warning to place governmental concerns into the application record. The GAC used consensus to activate the Guidebook’s strongest presumption. The Board used that presumption to direct that processing stop. Reconsideration tested a narrow category of informational error and left the result intact. The IRP tested compliance with ICANN’s governing duties, declared the earlier decision method inconsistent and recommended renewed independent judgment.

The Board then sought information, accepted negotiation-focused GAC advice, directed facilitation, removed the stop status and eventually accepted a commitments package. ICANN organisation processed the applications and executed contracts. The IANA functions verified delegation readiness and implemented the root-zone end state.

At no stage did mere participation confer final control. Governments participated early and powerfully, but the Board changed status. Amazon participated throughout and eventually prevailed in review, but the Board still decided whether to reopen processing. ACTO obtained contractual standing, but not registry management. The IRP exercised review authority, but not implementation authority. IANA implemented delegation, but did not adjudicate the cultural dispute.

The chain also shows how an institution can exercise public-like power through a private programme. ICANN was not allocating land or adjudicating territorial sovereignty. It was deciding access to a globally unique namespace under an application guidebook, bylaws, contracts and root-zone procedures. Yet the practical stakes—exclusive control of a meaningful label, recognition of governmental claims and the ability to exclude others—required disciplines associated with public decision-making: reasons, consistent standards, review and remedies.

Consensus advice was legitimate as a channel for governmental knowledge and public-policy risk. Its legitimacy weakened when the deciding institution treated consensus as an answer rather than evidence demanding a decision. The IRP did not diminish governmental participation by treating reasoned Board judgment as necessary. It preserved a distinction essential to the multistakeholder model: an advisory body may strongly influence a decision without silently becoming the decision-maker.

A counterfactual procedure would have separated evidence, judgment and bargaining

A better process would not have required ICANN to ignore governments or to accept Amazon’s applications automatically. It would have made the allocation of authority visible at the time of decision rather than after three years of review.

First, consensus advice that an application should not proceed should have opened a time-bounded public-policy evidence file. The GAC or the governments supporting the advice would identify the protected interests, the factual basis, the anticipated harm, the connection between the string and that harm, and the remedy sought. Where confidentiality was necessary, the Board could receive protected material while publishing a sufficiently detailed summary. The purpose would not be to force governments into courtroom pleading.

It would be to ensure that the strong presumption attached to a stated proposition rather than to consensus in the abstract.

Second, the applicant should receive a defined period to answer that file, with a right to submit evidence and proposed remediation. The response should be circulated to the body whose advice created the presumption, not merely posted elsewhere in the programme. If the GAC maintained its advice, it should state whether the response failed because the harm was non-remediable, the commitments were limited public evidence or the underlying claim concerned a categorical rule.

Third, the Board should publish a balancing decision. It should identify the programme rules, the weight accorded to consensus, the contrary findings from evaluation and objections, the legal analysis, the applicant’s undertakings and the specific reasons for accepting or rejecting rebuttal. The decision should distinguish claims the Board adopts from claims it merely records. That would make transparent participation answerable through accountable judgment.

Fourth, if commitments could cure the concern, negotiation should proceed under a fixed timetable and a disclosed menu of enforceable forms. The Board should specify whether the possible outcome is name reservation, eligibility restriction, shared registration access, a governance committee, information rights, audit, a third-party complaint mechanism or a combination. Each form allocates different power. Calling all of them “public-interest commitments” conceals more than it explains.

Fifth, failure to reach unanimity should return the matter to the Board by a date certain. A negotiation process should not give either side an indefinite veto unless the programme expressly confers one. The Board should decide whether the remaining risk justifies denial, conditional approval or a redesigned commitment. That decision should be reviewable under a standard capable of testing reasoning, not only missing information.

This counterfactual would not guarantee agreement. It would reduce the legitimacy deficit by ensuring that governmental input, Board judgment, negotiated conditions and implementation occurred in recognisable stages. It would also shorten the period in which application status, diplomatic negotiation and accountability review pulled in different directions.

The bounded conclusion

The .AMAZON applications were not stopped because governments controlled the root zone. They were stopped because ICANN’s programme assigned exceptional weight to governmental consensus and the Board used its own authority to translate that weight into an operational status. The IRP did not reverse that chain by awarding the strings. It declared that the Board had to use the authority it had previously allowed the presumption to obscure and recommended a fresh decision.

Recovery was controlled by the Board because the practical institutional consequence of review was a renewed decision obligation. The Board could ask for reasons, define the negotiation period, remove Will Not Proceed, accept contractual safeguards and direct processing. Contracting then converted selected public-policy concerns into obligations with named beneficiaries, standing and remedies. Delegation followed only after a separate implementation check.

The outcome vindicated neither a theory of automatic corporate entitlement nor a theory of governmental ownership. Amazon obtained the three delegated top-level domains. ACTO and its member states obtained enforceable, though bounded, contractual protections. The public record does not establish that every regional concern was satisfied, and it does not show that present use has produced the social or economic effects predicted by either side.

What the case establishes is narrower and more durable. Consensus can be strong enough to stop an application without being lawful as a substitute for judgment. Review can invalidate a process without delivering the asset at issue. Transparency can disclose a commitment while accountability depends on who has standing and what remedy follows. And in a distributed control chain, the institution that owns recovery is usually the one that can change the next operative status—not the one that first raised the objection, and not the one that later declared the process defective.