Restricted reference database distributing matching signals across participating platforms

GIFCT Hash-Sharing Database

The GIFCT Hash-Sharing Database is a shared collection of content fingerprints and associated labels used to help participating services identify qualifying terrorist and violent extremist material. Members contribute signals, and GIFCT can also create hashes. Access and inclusion are governed separately from a platform's own enforcement decisions.

Admission filters and withdrawal visibility

The published taxonomy describes separate designation, qualifying-incident and behavioural routes. TCAP-sourced inputs must also satisfy GIFCT's parameters rather than entering automatically. GIFCT's transparency accounts report label corrections, hash removals and a nonviolent music-video removal. A contributor withdrawal becomes visible to another member at its next access, but that does not demonstrate the recipient processed it or revisited an earlier action. The correction examples are not a random database sample and cannot yield a database-wide error rate.

What the records show

GIFCT HSDB taxonomy — December 2022 v1.0GIFCT Hash-Sharing Database

defines qualified inclusion routes

UN list, qualifying incident and behavioral routes have separate predicates. Behavioral classification needs more than an ideological label. December 2022 terminology remains dated.

GIFCT Hash-Sharing DatabaseAltitude — inspected public implementation

is a separately credentialed source described by Altitude

2024-08-08 – 2024-08-08

README requires obtaining GIFCT credentials. No exact GIFCT adapter payload or actual deployment was inspected; do not inherit TCAP payload semantics.

Terrorist Content Analytics PlatformGIFCT Hash-Sharing Database

supplies URL-hash inputs subject to GIFCT’s own parameters

Taxonomy names TCAP but requires GIFCT inclusion parameters too. This filter defeats an inference that every TCAP label automatically enters HSDB.

Global Internet Forum to Counter TerrorismGIFCT Hash-Sharing Database

can create qualifying hashes itself

Annual report describes GIFCT hash creation as well as member contributions; no item-specific instance asserted.

GIFCT member technology companiesGIFCT Hash-Sharing Database

reported correcting labels and removing hashes

Dated 2022/2025 report accounts, not a verified universal correction service-level agreement.

2022 account describes review/sample corrections and a separate nonviolent music-video removal. Another member sees contributor withdrawal at its next access. Nonrandom feedback does not establish a database-wide error rate; no recipient processing or restoration proved.

Global Internet Forum to Counter TerrorismGIFCT Hash-Sharing Database

reported taking management and withdrawing five legacy accesses

2022 report period; individual withdrawal dates unspecified.

2022 report says GIFCT took full management and rescinded access from five legacy nonmembers that did not pursue membership. Historical organizational access change, not current membership count or content removal.

Further reading

GIFCT: organizational authority, financing and signal correction

Altitude: product governance and correction boundaries

Read the original sources 6

What the connections say

8 relationships
1

Global Internet Forum to Counter Terrorismreported taking management and withdrawing five legacy accessesGIFCT Hash-Sharing Database

2022 report period; individual withdrawal dates unspecified.

Reported by the cited source

2022 report says GIFCT took full management and rescinded access from five legacy nonmembers that did not pursue membership. Historical organizational access change, not current membership count or content removal.

Read the original source 1
7

GIFCT member technology companiesreported correcting labels and removing hashesGIFCT Hash-Sharing Database

Dated 2022/2025 report accounts, not a verified universal correction service-level agreement.

Reported by the cited source

2022 account describes review/sample corrections and a separate nonviolent music-video removal. Another member sees contributor withdrawal at its next access. Nonrandom feedback does not establish a database-wide error rate; no recipient processing or restoration proved.

Read the original sources 2