A guest post on RIPE Labs by security researcher Sasha Romijn, published alongside RIPE NCC's own disclosure response, reveals that the organisation's single sign-on session cookie was architecturally available to every HTTPS-serving host under the *.ripe.net domain — a scope that encompassed well over a thousand nodes outside RIPE NCC's direct control.
The cookie in question, crowd.token_key, authenticates users across the full suite of RIPE NCC services: the RPKI Dashboard, the RIPE Database, the LIR member portal, RIPE Atlas, resource-transfer workflows, and stored API keys. Romijn confirmed the token carried no binding to IP address, device, or browser, meaning a copy obtained from one session could be used freely on another network entirely.
- Cookie scoped to *.ripe.net, reaching 1,000+ Atlas anchor hosts at time of report
- mtg.ripe.net CAA record fixed within ~10 days of the February 2025 report
- anchors.atlas.ripe.net CAA record fixed before May 2025
- Bounty paid: €1,100 (critical tier), released 11 months after the report
- No re-authentication required to add admin users or create RPKI API keys
Two concrete attack surfaces were identified. During RIPE meetings, attendees on the conference Wi-Fi receive DHCP hostnames under mtg.ripe.net. Because no restrictive Certificate Authority Authorization record existed on that subdomain, any attendee could obtain a legitimate TLS certificate from a public CA for their assigned hostname, serve a page over HTTPS, and collect crowd.token_key cookies from any logged-in visitor who followed a link to that address. RIPE meetings draw the very operators who manage routing policy for Europe, the Middle East, and Central Asia.
The second surface was the Atlas anchor network. At the time of the report, more than 1,000 anchor nodes were hosted by external organisations under anchors.atlas.ripe.net, again without a restricting CAA record. A single rogue employee at any of those organisations could have exploited the same approach — the anchor's hosting organisation did not need to be compromised as a whole.
What makes stolen sessions more dangerous than the CSRF vulnerabilities Romijn also reported is the difference between blind writes and full authenticated access. The CSRF chain allowed an attacker to push changes to RPKI Route Origin Authorizations and RIPE Database objects without reading responses. A captured session token provides unrestricted read and write access, and can be extended: adding new admin-level users or creating RPKI API keys requires no further authentication and generates no automatic notification to the account holder or designated company contact.
Certificate Transparency logs show no certificates were actually issued for mtg.ripe.net addresses beyond an infrastructure record, and anchor-domain certificate activity predates the vulnerability period, suggesting neither vector was exploited.
RIPE NCC's own post-mortem acknowledges that the initial CSRF fix was incomplete — a subsequent review found the original proof-of-concept still worked — and that communication with Romijn fragmented across channels, slowing coordination between internal teams. The bounty payment arrived 11 months after the report. RIPE NCC states it is now overhauling its authentication and authorisation architecture and improving cross-team coordination for complex multi-service disclosures.
Romijn's recommended mitigations go beyond the two CAA records that were deployed: move Atlas anchors and probes off *.ripe.net entirely, restrict CAA records to specific account URIs so that domain control alone is insufficient to obtain a certificate, and require step-up authentication before critical account changes such as adding users or rotating API keys.
For hosting and infrastructure operators running SSO across subdomains that include third-party or user-controlled hostnames, the case illustrates that cookie scope and responsible-disclosure scope can diverge in consequential ways: RIPE NCC's bug bounty policy explicitly excluded the Atlas and meeting-network subdomains, yet the SSO trust boundary included them.
Automated pipeline · Security
Synthesized from 2 source items across 1 industry feed on 13 Jun 2026. Passed independent editor verification before publication. Style guide v1.1.
Sources
Decision trail
- Checking for duplicates — Duplicate story same-story cluster; write with candidate 5; cluster_primary=5
- Checking for duplicates — New story First coverage of RIPE NCC authentication cookie vulnerability affecting session tokens.
- Writing the article — Draft created article_id=3 slug=ripe-ncc-sso-cookie-scope-exposed-session-tokens-to-1-000-plus-third-parties
-
Editor review — Approved
- Factual grounding: Minor: The article states the bounty covered 'both this and the XSS+CSRF chain to RIPE Database' per the source, but the draft implies the €1,100 bounty was solely for the cookie-scope issue. Source 2 states the Intigriti report 'paid €1,100 (tier 1, critical) covering both this and the XSS+CSRF chain to RIPE Database.' The Key Facts block says 'Bounty paid: €1,100 (critical tier)' without noting it covered multiple issues — potentially misleading but not a fabrication. Minor issue.
- Factual grounding: Minor: The article describes RIPE NCC's post-mortem as acknowledging 'communication with Romijn fragmented across channels, slowing coordination between internal teams.' Source 1 does say communication became fragmented across multiple channels, which is accurate. No issue.
- Factual grounding: Minor: The article says 'Certificate Transparency logs show no certificates were actually issued for mtg.ripe.net addresses beyond an infrastructure record.' Source 2 says 'there are no certificates other than nscache.mtg.ripe.net.' Characterizing nscache.mtg.ripe.net as 'an infrastructure record' is a reasonable paraphrase but adds a slight interpretive gloss not explicitly stated in the source. Minor.
- Factual grounding: Minor: The article states 'anchor-domain certificate activity predates the vulnerability period.' Source 2 says 'there were numerous certificates around 2020, but these most likely represent legitimate certificates. No certificates have been requested for several years.' The draft's phrasing 'predates the vulnerability period' is a reasonable inference from the source but adds a slight interpretive framing. Minor.
- Style compliance: Minor: Word count appears to be approximately 720-740 words in the body, which exceeds the 620-word target and approaches the 750-word hard maximum. Editors should verify final count.
- Quote integrity: No blockquotes used as attributed quotes — only a Key Facts block. No issue.
- No copied phrasing: Minor: The phrase 'blind writes that could modify ROAs and RIPE Database objects, but could not read responses' is close to Source 2's 'blind writes that could modify ROAs and RIPE Database objects, but could not read responses.' The article uses 'The CSRF chain allowed an attacker to push changes to RPKI Route Origin Authorizations and RIPE Database objects without reading responses' — adequately paraphrased.
- Factual grounding: Minor: The standfirst mentions 'meeting-network laptops' as potential harvesters. Source 2 describes the attacker as 'anyone at the meeting, sitting in the audience with a laptop on the wifi.' The standfirst's framing that meeting-network laptops 'could silently harvest' tokens is accurate in that any HTTPS host under that subdomain could receive cookies, but the standfirst conflates the attacker's device with general attendee laptops in a potentially confusing way. Not a factual error but a framing issue.
- Assigning hero image — Pexels pexels_id=17323801
- Linking related stories — Linked 0 relations from 0 candidates
- Linking related stories — Linked 0 relations from 0 candidates
- Publishing — Published ripe-ncc-sso-cookie-scope-exposed-session-tokens-to-1-000-plus-third-parties

Discussion · coming soon
Be the first to join the thread when community discussion launches.