You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
graphql-java includes its own copy of guava. On master, the version of guava was recently updated from 32.1.2-jre to 33.7.2-jre because of CVE-2026-102554: #4498
Various security scanner tools identify the version of Guava that is included inside of the graphql-java jar file and report this associated CVE, raising alarms for users who use and deploy applications having the graphql-java jar.
Similarly to #4498, I don't think there is a security risk here for graphql-java. Reading the advisory carefully, I think that an attacker needs two things to attack an application:
they can supply data that ends up as keys in a HashBiMap, or as values in a LinkedHashMultimap.
they can control the hashCode() of that data (in practice, this means supplying strings or other value-based types).
Here, graphql-java does ship the vulnerable classes in its JAR, but those collections always use objects like GraphQLDirective, Vertex or other graphql-java types. The attacker cannot trigger hash collisions here. An application could be in theory vulnerable if it uses relocated classes under graphql.com.google.* directly in its codebase in a vulnerable scenario - I would argue this is both unlikely and a mistake in the first place.
This is good to know and thank you. I am trying to be careful not to suggest that graphql-java has any security vulnerability. I am happy to let someone else do such investigation. Hopefully no one considers any possible attack vector here, especially with the vulnerable classes in question being present.
Would it be possible to push out a new minor or service release when things like this come up? It's better to preempt and quash any concerns that relate to a high severity security vulnerability in a dependency. Otherwise, the burden is on other projects to explain or prove why their application is not affected by the version of guava that is embedded inside of graphql-java.
Because Guava classes and metadata are embedded inside graphql-java, enterprise scanners (Snyk, OWASP, Xray, Prisma) constantly trigger alerts. Downstream developers are stuck because standard Maven/Gradle dependency management can't reach inside the JAR to patch or exclude it.
Moving away from shading - either by replacing it with native Java features or making it a standard dependency - would eliminate these blocking alerts and free maintainers from cutting new releases every time a Guava CVE drops.
Reacted by benken-parasoft, Ainur, mephisto2120, Ulrich Grepel, Richard Miller, Dmitriy, Alex Cusworth and Anastasiya
We just released 24.5, 25.2 and 26.2 to address this issue.
Moving away from shading - either by replacing it with native Java features or making it a standard dependency - would eliminate these blocking alerts and free maintainers from cutting new releases every time a Guava CVE drops.
While we appreciate this advice and we agree in general that this would solve this problem more fundamentally, we don't plan to do either at the moment. We leverage Guava to use the more performant and more feature rich collections vs standard Java ones and they served us well over the years. We can't have it as standard dependency as this would mean GraphQL Java will cause countless dependency conflicts and there are no native Java collections with the same features. But we continue to evaluate options and we might change our approach in the future.
Activity
bclozel commentedon Nov 3, 2026
Similarly to #4498, I don't think there is a security risk here for graphql-java. Reading the advisory carefully, I think that an attacker needs two things to attack an application:
HashBiMap, or as values in aLinkedHashMultimap.hashCode()of that data (in practice, this means supplying strings or other value-based types).Here, graphql-java does ship the vulnerable classes in its JAR, but those collections always use objects like
GraphQLDirective,Vertexor other graphql-java types. The attacker cannot trigger hash collisions here. An application could be in theory vulnerable if it uses relocated classes undergraphql.com.google.*directly in its codebase in a vulnerable scenario - I would argue this is both unlikely and a mistake in the first place.benken-parasoft commentedon Nov 3, 2026
This is good to know and thank you. I am trying to be careful not to suggest that graphql-java has any security vulnerability. I am happy to let someone else do such investigation. Hopefully no one considers any possible attack vector here, especially with the vulnerable classes in question being present.
Would it be possible to push out a new minor or service release when things like this come up? It's better to preempt and quash any concerns that relate to a high severity security vulnerability in a dependency. Otherwise, the burden is on other projects to explain or prove why their application is not affected by the version of guava that is embedded inside of graphql-java.
datagitlies commentedon Nov 9, 2026
Because Guava classes and metadata are embedded inside
graphql-java, enterprise scanners (Snyk, OWASP, Xray, Prisma) constantly trigger alerts. Downstream developers are stuck because standard Maven/Gradle dependency management can't reach inside the JAR to patch or exclude it.Moving away from shading - either by replacing it with native Java features or making it a standard dependency - would eliminate these blocking alerts and free maintainers from cutting new releases every time a Guava CVE drops.
andimarek commentedon Nov 13, 2026
We just released 24.5, 25.2 and 26.2 to address this issue.
While we appreciate this advice and we agree in general that this would solve this problem more fundamentally, we don't plan to do either at the moment. We leverage Guava to use the more performant and more feature rich collections vs standard Java ones and they served us well over the years. We can't have it as standard dependency as this would mean GraphQL Java will cause countless dependency conflicts and there are no native Java collections with the same features. But we continue to evaluate options and we might change our approach in the future.