Platform Audit Approach

The periodic audit of Synapse activity is intended to surface potential threat scenarios concerning the privacy and security of data held in the Synapse. The approach to this audit is informed by an assessment of risks to priority data, such as the data sets associated with Synapse projects marked with restricted access control lists. The risk assessment process considers access control at the point when access is granted, when access is used, and when access may become uncontrolled.

Auditing may be done by analyzing a representative sample of activity or a comprehensive report of activity over the audit period. A comprehensive report is preferred when the queries driving the report can be targeted to precisely address the threat scenario. Sampling is used as an alternative when comprehensive reporting is not feasible to address a given audit query, such as for activity common to all users within the application.

Overview

The Synapse audit should occur twice a year, once in June and once in January. Each audit should contain data from the last two quarters. The purpose of the audit is to ensure that there have not been any data breaches or security risks during the respective audit period.

The Synapse Governance Team should submit the audit report annually during the Synapse continuing review, which occurs in October.

Timeline

In May and November, the Synapse Security Engineer should pull MD5 duplicate data, state change data, and top downloader data from the previous two quarters. The Security Engineer should post this data on Synapse, and email it to ACT@sagebionetworks.org.
During the months of June and December, the Synapse ACT Team will sort the state change and MD5 data and email respective community leads to determine whether any files pose a security risk. If the project owner is external to Sage, email Xa to investigate before emailing the external Synapse user.
The Governance and Security teams will draft an audit report following this example from 2020.
The Director of Governance will review and approve the audit report draft
Add in WIRB and security compliance submission details

To be done:

Breach SOP, where to store annotated files and completed reports, audit report template

Threat Scenarios

Data access

Synapse implements an access control system based on the properties of the profile attempting to access information and on properties of the data set itself. An account must be validated in order to access controlled use data, and an access restriction must be in place on data with a controlled use classification in order to implement this access control.

Threat: A Synapse user intentionally or inadvertently accesses controlled data without qualification of their account

Identify through data warehouse query and end user reporting:

Users who have posted or access controlled data without the the appropriate validation property on their account.
Users who should have access removed at a prior time no longer have access

Threat: A Synapse user with significant access to data intentionally or inadvertently shares access

Identify through data warehouse query and end user reporting:

A single file downloaded multiple times by a single user

Associated queries: MD5 duplicates, Restriction change of state, Top downloaders

Data handling

Synapse allows end users to upload data once they have certified their account through a training module. The certification process is an administrative control that trains users on appropriate data handling procedures. Once granted data upload rights, an end user is expected to respect the permission sets associated with the data sets they handle.

Threat: A Synapse superuser intentionally or accidentally copies or uploads a controlled data set without appropriate access controls

Identify through data warehouse query and end user reporting:

PHI accidentally/intentionally released without appropriate conditions
Original terms of data contribution are not respected. Data proliferated into Synapse beyond the original terms of use
Public Synapse spaces contain only data classified as public

Associated queries: MD5 duplicates, Restriction change of state

Data loss

A Synapse account may be permitted to access many data sets of differing classifications. An incident of account sharing or account compromise may result in the download of a data set beyond what is intended according to an access restriction.

Threat: A Synapse account with extensive access to controlled data sets may be compromised:

Identify through data warehouse query and end user reporting:

Detecting the exfiltration of data from Synapse correlated with large-scale download activity by a user

Associated queries: Restriction change of state, Top downloaders

Audit Constraints

The Synapse audit approach was revised in 2020 to focus on specific threats identified through a risk assessment process. Automated queries were designed to report on the activity related to each threat.

The audit reports are limited by the time spans available to the automated queries. Some queries are based on changes to properties of objects and a query may not be able to compare an event with activity outside of its observation window. In these cases, the query will not surface a conflict between the event and a prior state.

Data warehouse queries, documentation, and handling

Restriction change of state

#select t1.ID, t1.IS_CONTROLLED, t1.IS_RESTRICTED, t1.IS_PUBLIC, t2.IS_CONTROLLED, t2.IS_RESTRICTED, t2.IS_PUBLIC
select t1.*, t2.*
from (
    select ns2.*
    from NODE_SNAPSHOT ns2
    join (
        # most recent snapshot
        select ns1.ID, max(ns1.TIMESTAMP)
        from NODE_SNAPSHOT ns1
        group by ns1.ID
    ) nsmax1 on nsmax1.ID=ns2.ID
) t1
join (
    select ns2.*
    from NODE_SNAPSHOT ns2
    join (
        # snapshot a month ago
        select ns1.ID, max(ns1.TIMESTAMP)
        from NODE_SNAPSHOT ns1
        where ns1.TIMESTAMP < unix_timestamp('2019-09-01 00:00:00')*1000
        group by ns1.ID
    ) nsmax1 on nsmax1.ID=ns2.ID
) t2 on t2.ID=t1.ID and t2.VERSION_NUMBER=t1.VERSION_NUMBER
where not (t1.IS_PUBLIC = t2.IS_PUBLIC and t1.IS_CONTROLLED = t2.IS_CONTROLLED and t1.IS_RESTRICTED = t2.IS_RESTRICTED)
limit 100
;

Export the Restriction change of state table to a spread sheet to create a pivot table summary of the number of access control changes by project. Include this summary in the report.
Contact the project owner or community manager of each different project on the list to notify them that their files have been identified as anomalies through a regular Synapse audit.
For any responses that indicate inadvertent or inappropriately permissive access control changes, create a ticket within the Governance Jira space for investigation of a privacy incident.

Top downloaders

# top 20 downloaders by count(filehandle_id)
select fhdr.USER_ID, count(*) as c
from FILE_HANDLE_DOWNLOAD_RECORD fhdr
where fhdr.TIMESTAMP between unix_timestamp('2019-07-01 00:00:00')*1000 and unix_timestamp('2019-09-10 00:00:00')*1000
group by fhdr.USER_ID
order by c desc
limit 20;

Contact the account holder of each account returned by this query with a prompt like the following:
Your Synapse account has been identified during a routine Synapse audit as having accessed a large number of files in the last six months. This activity may be expected due to how you use Synapse, or may be the result of a compromised or shared account.
Please reply to this email message to confirm that you are not aware of a breach of your Synapse credentials and that you have not shared them with anyone else.
Summarize the responses for the report.
For any responses that indicate loss of control of account credentials, create a ticket within the Governance Jira space for investigation of a privacy incident.

MD5 duplicates

create table auditdb.fhd_detail2 as
select ls.ID, ls.VERSION_NUMBER, ls.NAME,
       ls.PROJECT_ID, ls.PARENT_ID, ls.BENEFACTOR_ID,
       ls.IS_PUBLIC, ls.IS_CONTROLLED, ls.IS_RESTRICTED,
       ls.FILE_HANDLE_ID, md5c.CONTENT_MD5, md5c.c as DUP_COUNT
from auditdb.latest_snapshot_202003 ls
join warehouse.FILE_HANDLE_RECORD fhr on fhr.ID=ls.FILE_HANDLE_ID
join auditdb.fhr_md5_count md5c on md5c.CONTENT_MD5=fhr.CONTENT_MD5
where ls.IS_PUBLIC=1 and (ls.IS_CONTROLLED=1 or ls.IS_RESTRICTED=1)
select fhdd.CONTENT_MD5 as MD5, fhdd.PROJECT_ID as SOURCE_PROJECT, fhdd.ID as SOURCE_ID, fhdd.VERSION_NUMBER as SOURCE_VERSION, fhdd.IS_PUBLIC as SOURCE_IS_PUBLIC, fhdd.IS_CONTROLLED as SOURCE_IS_CONTROLLED, fhdd.IS_RESTRICTED as SOURCE_IS_RESTRICTED,
       ls.PROJECT_ID as DUP_PROJECT, ls.ID as DUP_ID, ls.CREATED_BY, ls.IS_PUBLIC as DUP_IS_PUBLIC, ls.IS_CONTROLLED as DUP_IS_CONTROLLED, ls.IS_RESTRICTED as DUP_IS_RESTRICTED
from auditdb.fhd_detail2 fhdd
join FILE_HANDLE_RECORD fhr on fhr.CONTENT_MD5=fhdd.CONTENT_MD5
join auditdb.latest_snapshot_202003 ls on ls.FILE_HANDLE_ID=fhr.ID
where fhdd.FILE_HANDLE_ID <> fhr.ID and ls.IS_PUBLIC = 1 and (fhdd.IS_CONTROLLED <> ls.IS_CONTROLLED or fhdd.IS_RESTRICTED <> ls.IS_RESTRICTED)

Export the project summary table from the MD5 duplicates table
Contact the file owner on the list to notify them that their files have been identified as anomalies through a regular Synapse audit.
For any responses that indicate proliferation of files beyond intended, create a ticket within the Governance Jira space for investigation of a privacy incident.