Summary
- Both of AMS-IX Amsterdam's route servers use autonomous system number 6777, and the registry object for AS6777 says the number facilitates the route servers and "does not originate any prefixes on its own".
- The routing security a participant gets is one of four documented modes it selects, and the fourth applies no filtering at all — on a page that also claims MANRS compliance and warns that the validity of the IRRDB/RPKI data behind the filtering "is not guaranteed in any way".
The number both boxes answer to
AMS-IX publishes the table itself: Route Server 1 is rs1.ams-ix.net at 80.249.208.255 and 2001:7f8:1::a500:6777:1, Route Server 2 is rs2.ams-ix.net at 80.249.209.0 and 2001:7f8:1::a500:6777:2, both on ASN 6777, both running BIRD2 (AMS-IX route-server documentation).
The same page is explicit about what these two machines are not: "The route servers do not participate in the forwarding path, so they do not forward any traffic." They are passive, they never initiate a BGP session, and a participant opens the connection to TCP port 179. Every prefix handed over carries a large community of the form 6777:1900:0. Participants must announce the same set of prefixes to both, the master table runs to roughly 400K IPv4 and 110K IPv6 prefixes, and the recommended max-prefix is set at 400K and 110K accordingly.
So the number is a neighbour identity. In a BGP session the far end's AS number is what a router is configured against and what the session is authenticated as, and here one number covers two hosts. What that number is not — as far as anything published goes — is an origin.
What the register keeps about it
The RIPE Database object for AS6777 carries as-name AMS-IX-RS, the organisation ORG-AIEB2-RIPE (Amsterdam Internet Exchange B.V., Netherlands, org-type LIR), status ASSIGNED, created 2005-01-12T10:14:20Z and last modified 2024-06-06T13:21:20Z (rendering of the RIPE object, second rendering). Two maintainers guard it, RIPE-NCC-END-MNT and AMS-IX-MNT. Twenty-one years after registration the object still describes a service rather than a network, and it says so in two lines:
This AS facilitates the AMS-IX Route Servers. It does not originate any prefixes on its own.
RIPEstat reads the same number and tells the same story: name "AMS-IX-RS Amsterdam Internet Exchange B.V.", RIR RIPE NCC, registration date 2005-01-12, country NL, plus two verdicts — "Does not originate prefixes visible in RIS" and "Seen in RIS as transiting" (RIPEstat AS6777).
Transiting is the right word for this shape of network. It appears in paths on the way to somewhere else, and it is not the somewhere else.
One register, two printings: RIPEstat labels the status ALLOCATED, the aut-num object labels it ASSIGNED. Both are printed here and neither is adopted.
Then there is the object of a different type. IRRd3 counts exactly one route object attributed to AS6777 — 0 IPv4, 1 IPv6 — the peering-LAN prefix 2001:7f8:1::/64, source RIPE, and its own validator flags that object RPKI-invalid (IRRd3 AS6777). A route6 object is a statement an operator files in an IRR; it is not a BGP announcement and it is not an observation. I did not verify whether that prefix is announced. The remark about originating nothing and the existence of a filed object are two different kinds of statement, and the honest reading keeps them apart instead of scoring one against the other.
The filtering is a menu, not a wall
This is where the register stops being a curiosity. The same documentation sets out "3+1" peering modes: filtering on both IRRDB and RPKI data (the default for new sessions), IRRDB only, RPKI only, and a fourth option the page itself calls not recommended (AMS-IX route-server documentation):
Peering mode 'Just tagging': By selecting this not recommended option, no filtering is applied to announced prefixes.
What arrives instead of a filter is a set of tags, a large community per verdict: ROA VALID 6777:65012, INVALID 6777:65022, UNKNOWN 6777:65023, prefix present in the sender's announced AS/AS-SET 6777:65011, not present 6777:65021. The full table, labelled.
The default is not nothing. Incoming sanitation drops RFC 1918 space, bogon and Martian prefixes and the default route from a list based on Team Cymru's BOGON list, and rejects an announcement whose next hop points at somebody else's router. Amsterdam adds two filters of its own: ROA-status INVALID prefixes are dropped, and so are prefixes not announced in the sender's aut-num export statements. Prefix filters draw on seven official IRR sources — RIPE, ARIN, APNIC, AFRINIC, LACNIC, JPIRR, IDNIC — with NTT, LEVEL3 and RADB used at lower priority. Policy changes are re-detected hourly from midnight Amsterdam time.
BGP roles per RFC 9234 are supported with the exchange as rs_server; rs_client is expected but optional, and sessions without the role are still accepted.
And then, in the same document, two sentences that belong side by side:
AMS-IX Amsterdam Route Servers are MANRS (Mutually Agreed Norms for Routing Security) compliant.
Note that validity of the IRRDB/RPKI based information provided is not guaranteed in any way. Please consider carefully whether your AMS-IX facing router should solely rely on information exchanged to and from the route servers.
These are not in contradiction. One is a claim about the operator's own practices. The other is a statement about the data those practices consume. Read together they define an exchange that filters carefully and then tells you not to trust the thing it filters with — unusually candid, and the reason the mode matters. A participant who leaves the default on receives a filtered view with a stated warranty of zero. A participant who selects the tagging mode receives everything, unfiltered and labelled. Neither of them receives "the AMS-IX filter" as a fixed property, because no document I read describes such an object.
The roadmap sentence that outlived its intention
A 2018 AMS-IX presentation on RPKI and IRR filtering lists, among the exchange's own MANRS roadmap actions, making the tagging mode inaccessible — "Make "Just tagging" inaccessible" — alongside making IRRdb plus RPKI the default (2018 presentation). Eight years later the live page still offers the mode, still calls it not recommended, and still warns that a router should not rely solely on what it receives.
Two dated facts, held apart. One intention, stated at a conference in 2018. One page, undated, that still prints the option. Nothing follows from the gap except the gap: an intention is not evidence of an action, and an undated page is not evidence of a recent decision.
What this desk had already published
Three briefings on this exchange appear in the BTW archive, and the delta is worth stating exactly, because the same operator publishes most of what gets quoted about it. On 18 October 2026 a briefing read the connected-network count and found its unit undefined (cmvdx68qz04zzz26b7awqwwtm, slug ams-ix-903-connected-networks-unit-undefined). On 11 November 2026 another read the peak traffic figure against eight separate connectivity products (cmwbi13fb00003a6bjqargn74, slug ams-ix-peak-15-822-tb-s-ar).
On 12 November 2026 a third read fifteen colocation facilities and the cable the member buys for itself (cmwcthtdc000bys6bpkyvhbax, slug ams-ix-fifteen-buildings-one-cable). None of the three reads AS6777, the route-server register identity, the peering-mode menu or the registry object. That is the gap this briefing fills.
What I read, and what I did not
Numbered, because the limits are load-bearing:
- Everything here is a reading of documents retrieved on 18–19 November 2026. The retrieval limits are facts about access, not about AMS-IX, its members or the RIPE NCC.
- The RIPE REST endpoint for the aut-num object returned no readable text (rest.db.ripe.net AS6777), so the as-name, organisation, status, dates and remark wording come from independent renderings of RIPE whois output and are mirror-derived, not registry-verified.
- Those renderings disagree on contacts: two show redacted DUMY-RIPE handles and carry a notify field, one shows unredacted handles without it. Both states are reported here; neither is adopted.
- Status is printed two ways: RIPEstat ALLOCATED, aut-num ASSIGNED.
- One route6 object, 2001:7f8:1::/64, is attributed to AS6777; whether it is announced in BGP was not verified, and the "does not originate" remark is quoted as what the object says, not judged against it.
- The RPKI-invalid flag is IRRd3's own validation verdict, not a RIPE statement.
- The route-server page carries no visible publication or revision date, so its sentences cannot be dated.
- That page publishes no numeric route-server ASN count. The platform-wide ASNS counter shown elsewhere on the site was not verified as a route-server figure and is not used.
- The page defines no "connected network", and this article supplies no definition of its own.
- The 2018 presentation was seen as search-returned slide text, not as a full read of the deck; it records an intention stated then, and the live page remains the dated fact beside it.
- The Hong Kong and Eastern Caribbean route-server pages reproduce the same wording at excerpt level only; nothing is quoted from them.
- No BGP session was established, no looking glass or route-server peer was queried, no announcement was measured, and no RPKI repository or IRR database was searched.
- The route-server page was read twice in the same retrieval round; the two reads are consistent where they overlap, and the first was truncated before the filtering and mode sections.
- No participant is identified. Every statement about what a participant selects is a statement about what the documentation offers, not about what any network chose.
Nothing above asserts misconduct, misstatement or intent by AMS-IX, the RIPE NCC or any member.
The consequence
A network that connects to AMS-IX and takes the route servers is not buying a security posture. It is choosing one. Leave the default and you get a filtered feed with an explicit zero-warranty on the data behind the filter. Take the tagging mode and you get the whole table with labels and nothing removed except the obvious junk. The exchange's checkable identity in the register — one AS number, facilitating, originating nothing, seen in transit — is honest about being a service. The risk, and the choice, sit one layer out, with whoever points a router at rs1 and rs2.
Member Briefing
Read the full trend briefing
Members read the full briefing with the source notes behind every finding.
Strategic Circle
A daily briefing in your language, built from changes we have verified.
Join for USD 20/monthCancel anytime.
Members also get the full research library, a daily briefing on the networks they follow and the private forum.






