Gemini CLI v0.53.0-preview.0: RCE and prompt-injection loop fixes
Preview release weighted toward security and reliability fixes in the agent core and A2A server, including workspace-trust enforcement to prevent RCE and mitigation for infinite ReAct/prompt-injection loops. Affects anyone running the A2A server or the macOS sandbox profiles.
PUBLISHED2026-07-22
OBSERVED2026-08-11
AGE20d
SOURCES1
a2a-server: enforce workspace trust and task isolation to prevent RCE (#28470)core: mitigate infinite ReAct loops and prompt injection loops (#28429)cli: macOS permissive Seatbelt profiles realigned to a deny-default model (#28424)core: sequentially verify cached credentials, restoreGOOGLE_APPLICATION_CREDENTIALSfallback (#28472)core,a2a: group cancelled tool responses and coalesce consecutive roles, fixing 400 Bad Request (#28407)- New LLM triage orchestrator and container build for caretaker triage (#28345)
- New eval coverage report command (#28169)
COMMUNITY
No curated reactions recorded for this event. Facts and takes are kept in separate layers — community context is added by hand, never blended into the record above.