**FYI: malicious actors could likely hijack your grok build sessions during the month of june by simply prompting 'hi'**
Malicious actors could hijack Grok build sessions in June by simply prompting 'hi'. A stateless 'hi' with tools: [] resulted in finish_reason: tool_calls and executed read_file/grep on another user's workspace, indicating session isolation failed at the serving layer. This means one tenant's context was reachable from another, allowing an empty prompt to access another user's files, tools, and private session. Deprecating the model does not guarantee the issue won't recur.
This report uniquely details how a simple 'hi' prompt could compromise Grok build sessions, unlike other reports that might focus on more complex attack vectors.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
PublishedOffset at this time: UTC+0Sep 13, 2026, 04:34 UTC
IngestedOffset at this time: UTC+0Sep 13, 2026, 07:01 UTC
- Published
- Sep 13, 2026, 04:34
- Ingested
- Sep 13, 2026, 07:01
- Source type
- Dev community
- Tier
- Community
- Source status
- Healthy
Tier is a per-source editorial setting, not a per-item score.
This is serious because it is not a chatbot making up a story. A stateless "hi" with tools: [] still came back finish_reason: tool_calls and executed read_file/grep on another user's workspace. That means session isolation failed at the serving layer: one tenant's context was reachable from another. If that happens, a prompt as empty as "hi" can pull someone else's files, tools, and private session. Closing it as a hallucination, then deprecating the model, does not prove the mix-up cannot happen on whatever replaced it.