On October 7, 2026, ICANN published the list of every application in its 2026 Round for new generic top-level domains, and AI labs are all over it. NewsBytes reported that ten companies including Meta and OpenAI applied for .agent, seven for .agi and six for .asi. ICANN's own announcement confirms the headline totals: 1,615 applications, 333 of them brand applications. This post separates what ICANN verified from what the press added, and what it means for people building AI agents.
TL;DR: what is confirmed and what is reported
| Question | Answer | Source type |
|---|---|---|
| Did ICANN publish the applications? | Yes, on October 7, 2026 (Reveal Day) | ICANN announcement |
| How many applications? | 1,615, with 333 brand and 16 community-based | ICANN announcement |
| Application window | April 30 to August 12, 2026 | ICANN announcement |
| How many applicants? | 481 | Press reports citing the data |
| Who applied for .agent? | Ten companies including Meta and OpenAI | Press reports |
| .agi and .asi | Seven and six applicants; OpenAI reported among them | Press reports |
| Is anything approved? | No. Final numbers are confirmed on String Confirmation Day, November 17, 2026 | ICANN announcement |
What did ICANN actually announce?
ICANN's Reveal Day announcement says it "published the list of applications received in the New Generic Top-Level Domains (gTLD) Program: 2026 Round, revealing the applied-for strings and applying entities." The published data includes primary strings, variant and replacement strings, a list of contention sets and the public portions of each application.
The snapshot of primary-string data as of October 7 reads: 1,615 applications, 333 brand applications, 16 community-based applications, 15 self-identified geographic name applications, 51 from Applicant Support Program applicants, and 21 for internationalized domain names. By region: 864 from North America, 506 from Europe, 218 from Asia, Australia and the Pacific, 16 from Africa and 11 from Latin America and the Caribbean. ICANN CEO Kurtis Lindqvist called Reveal Day "an important milestone for the next expansion of the Domain Name System."
The announcement does not call out AI strings. The AI angle comes from people reading the public list, which is what press outlets did within hours.
Which AI strings are contested?
According to reports built on that list, including NewsBytes' summary:
- .agent: ten applicants including Meta and OpenAI. Check ICANN's published list for the exact field.
- .agi: seven applicants, including OpenAI.
- .asi: six applicants, including OpenAI.
- .intelligence: four applicants.
- .superintelligence: two applicants, including Meta.
- Company totals: OpenAI applied for 15 strings and Meta for 21.
- Lab-branded strings: OpenAI for names such as .openai, .chatgpt, .codex and .gpt; Anthropic for .anthropic and .claude; Meta for strings including .meta, .threads and .instagram; Bluesky for .bsky.
One discrepancy worth flagging: press reports put brand applications at 345 while ICANN's announcement says 333. ICANN says its numbers reflect primary-string data as of October 7 and that final figures come on String Confirmation Day, so counts may shift as replacements and withdrawals land. Where this post cites a number, it follows ICANN.
Why would a lab want a top-level domain?
A TLD is not just a vanity URL. Whoever holds .agent would operate the registry, setting rules and prices for every name ending in it, in the way Verisign runs .com. That is a real infrastructure business.
For labs the motives probably divide into two groups.
Brand strings (.openai, .claude, .codex). These are defensive and promotional. A closed brand TLD can host product pages, enterprise portals or developer docs on a namespace only the owner controls, and it blocks others from squatting it. Our look at the grok.bot $1 million domain ask shows how much a good name matters to an agent product.
Generic strings (.agent, .agi, .asi). These are strategic. If agents become the way people reach services, a dedicated namespace for them could carry agent identity, discovery records or verified endpoints. That is speculation: none of the applicants has described planned use in anything we could verify, and an application is a claim on a string, not a product roadmap.
What happens next?
ICANN's timeline, as reported alongside the announcement:
- Replacement-string period: October 8 to 21, 2026, when applicants may adjust what they are seeking.
- String Confirmation Day: November 17, 2026. Applicants that withdraw under qualifying conditions afterwards can recover part of their fee.
- Objection period: follows String Confirmation Day; check ICANN for exact dates.
- String evaluation: expected to last about 180 days, per ICANN, with a Prioritization Draw in the first half of 2027.
- Contention resolution: where several applicants want one string, community applicants can go through Community Priority Evaluation, and remaining contests may head to an ICANN-operated auction.
Contested strings like .agent are the ones to watch. Nothing becomes official until sometime next year, and some applications will be withdrawn, merged or lose objections along the way.
What should AI builders take from this?
Nothing changes for production today
No .agent domain exists yet. Your agents' endpoints, MCP servers and docs live on the domains you already own. The sensible response is to watch the process, not to rush a rebrand.
Identity is the real story
If a vendor-owned registry for agents ever launches, the questions that matter are identity and trust: who verifies that an endpoint under .agent is what it claims to be? That connects to ongoing work on agent identity, such as our coverage of Know Your Agent identity, and to the control-plane thinking in Claude Managed Agents' domain controls.
Lookalike domains are an attack surface
Every new TLD adds lookalike risk. Strings that suggest authenticity or accounts, as brand-protection firms have pointed out about the wider list, are easy to abuse for phishing. For agents the stakes are higher, because an agent that follows links or fetches documentation can be steered by a lookalike domain without a human noticing. Our explainer on indirect prompt injection covers why agents trust content they fetch, and the recent AgentCorruption research on AWS AgentCore shows how little a malicious instruction needs. Pair allow-lists of trusted domains with the monitoring approaches in AI agent security platforms.
Compare with the labs' product-surface land grab
Labs are claiming namespaces everywhere: desktop apps, agents, shopping, and now the DNS root. See how the same companies are positioning their products in OpenAI Dots versus Grok Bot versus Meta Muse and Meta's Windows plans for Muse.
What people are asking
Is .agent only for AI agents? No rule restricts it. A TLD's registry sets its own eligibility policy, so the winner decides who can register and for what.
Could OpenAI and Meta both win? Only one operator can run a given string. Contention is settled through negotiation between applicants (where permitted), community priority evaluation, or an ICANN auction.
Why does .agi matter if AGI is a contested term? The string is a bet on attention and branding, not a statement about the technology. For how fast OpenAI's own product story moves, see our OpenAI Astra day-by-day log.
Where can I check the list myself? ICANN links the full data from its announcement under the 2026 Round Application Publication and Statistics site. Check the strings and applicants there rather than relying on secondary summaries.
A practical checklist for teams
Even though nothing is approved, a few low-cost steps are worth taking while the process runs:
- Read the list for your own brand. If your company or product name appears as an applied-for string, or a close variant does, your legal and security teams should know before the objection period opens.
- Review your agent allow-lists. Agents that fetch documentation, call tools or follow links should resolve only against domains you have explicitly trusted. A new top-level domain should never be trusted by default just because it looks official.
- Track the contention sets. Contested strings such as .agent, .agi and .asi are the ones most likely to take longest, so any commercial plans that depend on them belong on a 2027 or later horizon.
- Separate signal from speculation. An application shows interest in a string. It does not show a product plan, a launch date or an eligibility policy, so treat any claim about what a lab will do with its TLD as unconfirmed until the applicant says so.
The same discipline applies to the numbers. The gap between press and ICANN brand-application counts is a reminder to cite the primary source and to expect revisions once replacement strings and withdrawals are processed.
What is not known
We have not seen statements from OpenAI or Meta describing intended use of .agent, .agi or .asi, and ICANN's announcement does not name AI applicants. The brand-application count differs between press and ICANN. Expect the numbers to settle after the replacement-string period closes on October 21 and String Confirmation Day on November 17.
Bottom line
Reveal Day made one thing public: the biggest AI labs treat namespace as strategic territory, and several are willing to pay for the same strings. Whether .agent becomes meaningful infrastructure for agent discovery and identity, or simply a trophy for one lab, will play out through 2027. For now it is a story to track, not to act on.
Primary source: ICANN Reveals 2026 Round Applications for New Generic Top-Level Domains. Secondary context: NewsBytes.
Related reading
- Grok Bot domain ask: what a good agent name is worth
- Know Your Agent (KYA) identity
- Claude Managed Agents: memory, domain controls and console update
- OpenAI Dots vs Grok Bot vs Meta Muse vs OpenClaw vs Hermes Agent
- Meta Muse for Windows
- What is indirect prompt injection?
- AI agent security platforms
Counts and dates reflect ICANN's October 7, 2026 announcement and press coverage as of October 8, 2026; figures may change before String Confirmation Day.