Grok can now be invoked directly in-thread on X, meaning users trigger LLM responses by tagging a bot rather than through an API or app—so any factual claim, code snippet, or reasoning it emits is instantly public and unversioned. If you're shipping anything that competes with or gets cited alongside these public outputs, expect users to treat casual bot replies as authoritative, and expect your own model's public-facing errors to be screenshotted and amplified the same way.