Thirteen companies filed to run the top-level domain .agent, more than for any other string in ICANN’s 2026 round of new internet suffixes, according to CircleID’s October 7, 2026 coverage of Reveal Day. Under the same round’s new rules, none of the twelve losers in that pile-up can be paid to walk away — the private settlements that resolved most contested strings in 2012 are explicitly banned this time.

The String That Beat Every Other Bid in the Round

The Internet Corporation for Assigned Names and Numbers (ICANN), the nonprofit that runs the domain name system, published its full list of 2026 Round applications at 18:00 UTC on October 7, 2026, according to ICANN’s own announcement and CircleID’s same-day report. The release covered 1,615 applications from 481 applicants — one fewer than the 1,616 ICANN had said would proceed as of its September 22 confirmation — spanning 968 unique strings and 263 contention sets, where two or more applicants want the identical suffix.

.agent topped all of them at 13 competing applications, ahead of .bit at 11 and a three-way tie between .api, .hub and .brand at 10 apiece, CircleID reported. That is a striking result for a round long expected to be dominated by generic commerce strings: the single most fought-over suffix in ICANN history is one built for software that acts on a person’s behalf, not for shops, brands or even artificial intelligence itself.

Who Is Actually Inside the .Agent Contention Set

The names behind the thirteen filings read like a list of who believes AI agents need their own namespace. CircleID identified OpenAI Opco LLC, which filed 15 primary applications including .agent, .agi and .mcp (a reference to the Model Context Protocol, the standard several AI vendors use to let agents call outside tools); Meta Registry Holdings LLC, which filed 18 primary applications including .agent alongside uncontested dotBrand strings like .facebook and .instagram; and Charleston Road Registry, the registry arm through which Google filed 39 applications, 13 of which face competition.

Rounding out the set, per CircleID, are the portfolio TLD operators XYZ.COM and Radix Technologies, which run dozens of generic suffixes as an ongoing business; a subsidiary of Identity Digital, the registry operator that separately took over the .us registry contract from GoDaddy this month; an entity whose public presence points to Alibaba Cloud; and Open Agent Registry, described in the coverage as a community bid representing an “Agent Community” constituency rather than a commercial registry.

What Banning the Payoff Actually Costs the Losers

In the 2012 round, 234 contention sets went to resolution and only 16 ended up at a public ICANN-run auction — the rest cleared because applicants negotiated private deals, often splitting a payout among the bidders who agreed to withdraw. The 2026 Applicant Guidebook closes that route entirely: private auctions, joint ventures and any arrangement that compensates a rival for dropping out are prohibited, leaving only a Community Priority Evaluation, where one qualifies, or an ICANN-run auction whose proceeds go to ICANN’s own reserves, not to the applicants who lose.

For .agent, that process is already moving. Contested strings undergo Community Priority Evaluation first, a review expected to take roughly 180 days, before any unresolved set proceeds toward auction, with a prioritization draw reportedly expected in the first half of 2027, according to TokenPost’s reporting on the round. Every one of the thirteen applicants has already paid ICANN’s base evaluation fee of roughly $227,000 per application, a floor that domain-industry trade press has noted often rises with extra review costs — money that, under the new rule, none of the twelve eventual losers gets back in any form.

The One Applicant Without a Trillion-Dollar Balance Sheet

For OpenAI, Meta, Google, Alibaba Cloud, XYZ.COM, Radix and Identity Digital, a lost bid for one string among dozens filed is a line-item write-off. Six of the seven run portfolio or dotBrand strategies spanning double-digit applications each; absorbing a six-figure sunk cost on .agent specifically changes nothing about their other filings.

Open Agent Registry’s position is structurally different. A single-string community applicant has, by definition, no other application to amortize the loss against, and no joint venture or buyout now available to recover any part of its outlay if a Community Priority Evaluation does not go its way. Its only route to the string is to win that evaluation outright or to outlast seven-figure, multi-application rivals in a public auction that pays the loser nothing. The rule this blog flagged in September as closing off backroom deals was framed then as a fairness fix aimed at speculative filers; the .agent contention set is the first concrete case showing who actually absorbs the downside once that exit is gone, and it is the applicant without a portfolio to spread the risk across, not the platforms with one.

Key takeaways

  • ICANN’s October 7, 2026 Reveal Day confirmed .agent as the most contested string of the 2026 new gTLD round, with 13 competing applications versus 11 for the next-closest string, .bit.
  • OpenAI, Meta and Google’s Charleston Road Registry are among the named applicants for .agent, alongside portfolio registries XYZ.COM and Radix, an Identity Digital subsidiary, an Alibaba Cloud-linked entity, and the community bid Open Agent Registry.
  • The 2026 Applicant Guidebook bans the private settlements that resolved most 2012-round contention sets, leaving only Community Priority Evaluation or a public ICANN auction that returns nothing to the applicants who lose.
  • Each of the 13 applicants has already paid ICANN’s roughly $227,000 base evaluation fee, a cost that portfolio registries and Big Tech filers can absorb across dozens of other applications but that a single-string community applicant cannot recover in any other filing.
  • Brand, licensing and tech-transfer teams evaluating a future TLD application should underwrite that fee as a cost with a real chance of total loss, not a negotiable deposit, now that the backroom buyout that softened that risk in 2012 no longer exists.