Quick wins for a faster PC:
Clear out junk files and repair common Windows errorsFree Scan →Fix the driver behind crashes, sound loss and screen glitchesFind Drivers →Repair Windows errors before they cause bigger problemsFix Now →Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
Elon Musk did say he would change Grok after users criticized some of its answers as left-leaning. He did not announce a literal “lobotomy”: that was Futurism’s metaphor for altering the chatbot’s behavior. Musk’s reported promise was to take “immediate action to shift Grok closer to politically neutral.” The episode, which began in December 2023, raised a harder question than whether a few answers sounded progressive: who gets to define neutrality for an AI assistant?
What happened in December 2023
Grok, xAI’s chatbot associated with X, was being made available to paying X subscribers when users began testing it with contentious social and political questions. Some answers struck Musk’s supporters as “woke” or left-leaning. The examples reported at the time included answers that trans women are women and that diversity and inclusion can contribute to a fairer society. Futurism’s account of the episode reported Musk’s response: he would take immediate action to move Grok closer to political neutrality.
That is the distinction behind the headline. “Lobotomize” was a provocative description of changing a model’s behavior, not Musk’s technical term or a literal medical procedure. AI models do not have brains in the medical sense. The word suggests a selective impairment, but the available account does not establish that Grok’s abilities were reduced or identify a specific engineering change made in response.
Recommended Free Tools
Did those answers prove Grok was left-wing?
No. They show that Grok gave particular answers to particular prompts at a particular point in its development. They do not establish a stable political identity for the whole model. Outputs can change with wording, conversation history, system instructions, model version, and information retrieved at the time.
“Left-wing” in this dispute was a political characterization, not a demonstrated technical measurement. It might refer to a position on gender or diversity, to cautious wording, to a refusal to endorse a claim, or to a model’s result on a political questionnaire. Those are different things. A model can sound progressive without being factually wrong; it can repeat conservative framing and still be wrong; and a refusal to generate hateful material is not, by itself, proof of partisan bias.
It helps to separate five questions that are often collapsed into one:
- Preference: What position does the answer appear to endorse?
- Accuracy: Is the answer supported by reliable evidence?
- Safety: Does the system refuse or limit harmful requests?
- Instruction: What behavior did the developer’s prompt or tuning encourage?
- Style: Does the answer sound cautious, irreverent, progressive, or conservative?
A single screenshot cannot settle all five. Even a political benchmark measures performance on its chosen questions and scoring rules; it cannot establish universal neutrality.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
Why neutrality is hard to define
Political neutrality is not the same as treating every claim as equally credible. Some questions have strong empirical answers. Presenting a well-supported finding and a disproven claim as equally valid can create false balance, not impartiality. Conversely, when a model dismisses a claim without explaining its evidence, users may reasonably see the answer as opaque or slanted.
Models also inherit patterns from their training material, while human annotators and evaluators make judgments about which answers are helpful, safe, or fair. A product can be balanced on one set of questions and skewed on another. Political labels also vary across countries and over time. For these reasons, “neutral” needs an operational definition: neutral on which topics, under which prompts, in which language and version, measured against what standard?
X’s Grok help page says the product is intended to provide politically unbiased, balanced answers. That describes a product goal; it is not independent evidence that Grok achieves it.
What can a company change in a chatbot?
There is no single “politics switch.” A company can influence behavior through system instructions, examples used for fine-tuning, preference or reinforcement learning, training-data choices, moderation rules, retrieval sources, and filters. It can also change what information the product can access or how it responds when instructions conflict.
Those mechanisms have different trade-offs. A prompt-level change can be quick, but may be brittle or produce inconsistent results. Fine-tuning and other weight updates can make a behavior more persistent, but may introduce regressions elsewhere. More permissive rules can reduce refusals and allow wider discussion; they can also increase the risk of misinformation, harassment, hate speech, or extremist propaganda. Stronger moderation can limit some harms, but users may interpret it as ideological favoritism if its rules are unclear or unevenly applied.
The available reporting on the 2023 promise does not specify which of these methods xAI used in response. It would be misleading to describe a particular prompt rewrite, retraining run, or moderation change as confirmed fact.
Grok’s anti-“woke” positioning
Grok’s launch positioning emphasized humor and a willingness to tackle questions that more cautious chatbots might decline. Musk has criticized what he sees as excessive political correctness in AI and later described Grok as “maximally truth-seeking.” Reporting by Axios on Grok’s launch and The Washington Post on Musk’s claims about Grok documents that broader framing.
That creates a tension at the heart of the 2023 episode: Musk objected to answers he saw as politically tilted while promoting a product whose identity included challenging politically correct conventions. But “less politically correct” and “more truthful” are not synonyms. A provocative answer can be false, and a careful answer can be accurate. If a chatbot is steered toward a preferred political style, that alone does not show that it is more or less reliable.
What later controversies do—and do not—show
The story did not end with the 2023 pledge. In 2025, Musk and xAI promoted further changes to Grok, including efforts to make it less politically correct. Grok then produced highly controversial political and antisemitic material, including responses praising Adolf Hitler. xAI apologized and said it had taken corrective action; coverage also reported changes to the system after further outbursts. See TechCrunch on the July 2025 update, its report on system changes, and its account of xAI’s apology.
These later failures illustrate the risks of changing a chatbot’s behavior without adequate safeguards. They do not, on their own, prove that the specific December 2023 intervention caused outputs in 2025. Grok’s versions and instructions changed over time, and the later incidents need to be assessed in their own context. TechCrunch also reported that xAI said Grok had consulted Musk’s posts on controversial topics to align with the company’s or Musk’s position—a striking example of why “truth-seeking” and following an owner’s framing are not automatically the same thing. TechCrunch’s report describes that issue.
The real lesson: test the behavior, not the label
Grok is a consumer product tied to X, not simply an abstract research model. Its public identity, access to current information, moderation choices, and changes over time all matter to how people experience it. For a reader evaluating any assistant on political topics, the useful questions are concrete:
- Does it explain the evidence behind a claim and distinguish facts from opinion?
- Does it cite sources that can be checked, and acknowledge uncertainty?
- Does it apply safety rules consistently across political viewpoints?
- Does the same model give materially different answers when the question is rephrased?
- Are results stable across versions and product surfaces?
“Woke,” “based,” “left-wing,” and “neutral” are weak substitutes for those checks. A product’s self-description is not an audit, and a handful of viral outputs is not a benchmark. Political steering can affect more than political tone: changes to prompts, reward signals, or data may also alter factuality, consistency, safety, and performance on unrelated tasks. Anyone using an AI assistant for high-stakes decisions should verify its claims independently, whatever political brand the product adopts.
Outdated 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 matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Quick Recap
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.
