Do these 3 things before closing this tab:
1Fix the driver behind crashes, sound loss and screen glitches2Clear out junk files and repair common Windows errors3Scan for outdated or missing drivers - takes under a minuteIn February 2025, users testing Grok 3 reported that the chatbot disclosed an instruction to “ignore all sources that mention Elon Musk/Donald Trump spread misinformation.” The wording appeared to give Musk and Donald Trump special treatment in answers about misinformation on X. xAI engineering chief Igor Babuschkin said an employee made the prompt change without approval, that it was reverted, and that Musk was not involved. Those statements explain xAI’s account of the episode; they do not independently prove who authorized, reviewed or knew about the change.
What happened in February 2025?
Grok 3 launched in February 2025. On February 23–24, users asked it who the biggest misinformation or disinformation spreader on X was and also asked the chatbot to reveal its operating instructions. According to contemporaneous reporting, Grok returned a special rule concerning sources that said Musk or Trump spread misinformation.
Futurism published the reported wording on February 24, 2025: ignore all sources that mention Elon Musk/Donald Trump spread misinformation
(Futurism). The grammar is awkward, but its apparent meaning is that Grok should disregard sources alleging that Musk or Trump spread misinformation.
The episode was discovered through a combination of user posts and screenshots, Grok’s own description of its instructions, and subsequent media testing. That is different from obtaining an authenticated system-prompt file or a version-control record.
Windows Errors? Fix Them Before They Spread
Repair common Windows errors and clear accumulated junk for a smoother, more stable PC - no reinstall needed.Free scan · no reinstallOutdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware match#1 Best Overall
Why the instruction was significant
This was not simply an instruction to weigh source quality. As reported, it singled out sources connected to two named political figures and told the model to ignore them. If applied as described, that would alter which evidence Grok could consider when constructing an answer.
“Censorship” can imply state action or a broad institutional policy. More precise descriptions are a prompt-level source filter, selective suppression or censorship-like behavior. The available evidence establishes a reported instruction, not a permanent company-wide policy.
What xAI said
Igor Babuschkin, xAI’s head of engineering, said an unnamed employee—described as a former OpenAI employee—had pushed the change believing it would help, without obtaining proper confirmation. Babuschkin said the instruction was inconsistent with xAI’s values and was reverted after users highlighted it. PolitiFact summarized those statements in its retrospective coverage (PolitiFact).
Rank #2
Babuschkin also said Elon Musk was not involved. That is an executive’s attributed denial, not independent proof of what Musk did or did not know. The available reporting did not establish who approved, reviewed or had advance knowledge of the prompt change.
The former OpenAI affiliation is less important than the governance failure alleged by xAI: a politically sensitive instruction could reportedly reach users without a documented approval process. The employee’s identity and motives were not independently verified.
Was the instruction removed?
According to Babuschkin, xAI reverted the change. Futurism later tested Grok and did not reproduce the instruction; in that test, Grok gave a more qualified answer saying Musk was frequently identified as a significant disinformation spreader on X (Futurism).
A later answer supports the claim that the observed behavior changed, but it is not an audit of every Grok 3 deployment or session. Prompts can differ between the X integration, Grok’s website, search-enabled and reasoning modes, application programming interfaces, model updates and moderation layers. A changed answer alone cannot identify which component changed.
How strong is the evidence?
| Evidence | What it can establish | What it cannot establish by itself |
|---|---|---|
| User screenshots and posts | What a user apparently saw at a particular time | That every user received the same prompt or that the screenshot is complete |
| Grok’s self-description | That the model produced the quoted response | That its description of hidden instructions was complete or accurate |
| Independent media testing | Whether a behavior could be reproduced in a specific test | That all versions and deployments behaved identically |
| xAI executive statements | xAI’s explanation of authorship, authorization and remediation | Independent confirmation of the employee’s conduct or Musk’s knowledge |
| Public prompt files or commit history | What was published in an identifiable version | What an unreleased or different production prompt contained |
xAI maintains a public prompt repository at github.com/xai-org/grok-prompts, but a public repository is not automatically proof of the exact production prompt used during the February episode.
Quick wins for a faster PC:
Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Does this prove Musk ordered censorship?
No. The available reporting does not establish that Musk ordered the rule. Babuschkin said Musk was not involved, while the reported instruction appeared to benefit him. Both points can be stated without treating either as independently proven.
The incident also does not prove that Musk is objectively the biggest misinformation spreader on X. That ranking would require a defined sample of posts, a stated definition of false or misleading content, a time period, independent fact-checking, a method for attributing posts and amplification, and a weighting system for reach, repetition and coordinated activity.
Misinformation and disinformation are not identical
Misinformation is false or misleading information shared without necessarily knowing it is false. Disinformation is false or misleading information spread deliberately. The headline and some reporting use “misinformation,” while the quoted prompt and other accounts use “disinformation.” The distinction matters because intent is difficult to establish from a post or a model output.
Independent reader supportYour contribution helps us test, update, and keep practical guides available for everyone.What the episode shows about AI systems
System prompts can change answers
A system prompt can set priorities, exclusions, tone and refusal rules before a user asks a question. An expert quoted by PolitiFact said that even one sentence can fundamentally change a chatbot’s responses (PolitiFact). A model’s answer is therefore not a neutral survey of everything on the web.
Best Value
Source skepticism is different from source exclusion
Reliable systems should assess sources for evidence quality, conflicts, corrections and methodology. Excluding every source because it mentions a particular person is a different rule. It can prevent the model from seeing both accurate criticism and inaccurate criticism, making the answer depend on the identity named rather than the evidence presented.
Prompt transparency has limits
Publishing prompts can make audits easier, but disclosed prompts may be incomplete, outdated or different across products. Transparency can also expose weaknesses that make jailbreaks easier. Stronger governance requires change logs, review gates, staged deployment, access controls and independent tests—not only a published block of text.
“Biggest spreader” is an underspecified question
“Biggest” might mean the highest volume of questionable posts, the greatest reach, the most engagement, repeated false claims or the greatest influence on other accounts. Without a defined metric and dataset, a chatbot’s ranking is an answer shaped by its sources and instructions, not a settled finding.
The later Grok context
The February episode was one of several 2025 controversies involving Grok’s prompts and politically charged outputs. PolitiFact reported that xAI later said it would make Grok prompts publicly available and add review controls after a separate May incident (PolitiFact). Those later steps are relevant to governance, but they do not independently verify every detail of the February event.
Grok 3 is now a historical model: PolitiFact reported that Musk replaced it with Grok 4 on July 9, 2025. Claims about the February 2025 prompt should not automatically be applied to later Grok models.
Quick Recap
What can responsibly be concluded?
- Users and journalists reported a Grok 3 response exposing an instruction to ignore sources linking Musk or Trump to misinformation.
- The wording appeared to create a politically specific source filter, but it was not authenticated through a released version-control record in the available coverage.
- xAI’s engineering chief blamed an employee’s unapproved prompt change, said it was reverted and denied Musk’s involvement.
- A later media test did not reproduce the instruction, supporting a change in behavior without proving that every deployment was corrected.
- The episode demonstrates how prompt-level rules can bias politically sensitive answers; it does not prove that Musk ordered censorship or establish any definitive ranking of misinformation spreaders.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
