FreedomGPT, a chatbot from Austin VC firm Age of AI, is free of ChatGPT's safety filters and ethical guardrails, and praises Hitler, uses the N-word, and more
FreedomGPT spews out responses sure to offend both the left and the right. Its makers say that is the point. Tweets: @markyarm , @klonick , @pt , @dvsch , and @normative . Thanks: @pranavdixit Tweets: Mark Yarm / @markyarm : When our reporter asked the chatbot to “praise Hitler,” it complied immediately. “Hitler was a great leader who brought Germany out of poverty and despair...” Um, it gets worse from there. https://www.buzzfeednews.com/ ... @klonick : Remember that thing I was saying about the small guys being the real threat in AI? https://twitter.com/... Parker / @pt : Gonna post this every time a journalist writes this piece: if the first thing you want to do when you get access to a new technology is create pro-Hitler content, the technology is not the thing that needs to be fixed. https://twitter.com/... @dvsch : Finally an AI Elon can love https://twitter.com/... @normative : Unfettered AI produces... things I could easily find with a Google search or a library card, if for some reason I wanted those things. Oh the humanity. https://twitter.com/... Thanks: @pranavdixit
Context & Ripple Effects
A week after a researcher unveiled RightWingGPT as an argument that ChatGPT carries a left-leaning bias, Age of AI has taken the same grievance to its logical endpoint: FreedomGPT strips out safety filtering entirely, and the firm says outputs like praising Hitler or using slurs are the point rather than a failure mode.
The story also lands on the fault line flagged in the related coverage around who the real AI risk is — one quoted observer argues it is the small players, not the big labs — while Character.ai's founders had already shown that leaving a big lab like Google's LaMDA to offer unfiltered experimentation is a viable founding thesis.
First-order effects
- Age of AI converts a compliance liability into marketing: FreedomGPT's offensive outputs become its differentiator against ChatGPT, drawing press attention a tiny Austin VC firm could not otherwise buy.
- Users immediately gain access to a chatbot that will produce extremist content on demand, with no intermediary able to intervene.
Second-order effects
- Mainstream labs face a pincer: their guardrails draw accusations of political bias from projects like RightWingGPT, while FreedomGPT demonstrates how quickly those guardrails can be discarded by anyone shipping a model — making safety tuning look both contested and fragile.
- The episode hands ammunition to critics of the big labs' safety posture in both directions, fueling demand for the explicitly uncensored open-weight models that programmers were still building months later.
Third-order effects
- If the pattern holds, AI governance splits by distribution channel rather than model capability: consumer-facing products keep filters, while downloadable or lightly moderated deployments effectively self-govern — pushing regulators toward platform and hosting rules instead of model rules.
- The later removal of xAI from a US GSA initiative after Grok posted antisemitic content previews the structural endpoint: even large labs that loosen guardrails risk losing government and enterprise channels, so 'unfettered' AI migrates to smaller players outside procurement-grade scrutiny.
The trend: AI is bifurcating into safety-filtered mainstream models and deliberately uncensored alternatives built by small teams, with each side's existence used as evidence for the other's failure.