OpenAI said in a new policy that it is changing the way it trains artificial intelligence models to explicitly champion "intellectual freedom... no matter how challenging or controversial a topic may be." As a result, ChatGPT will ultimately be able to answer more questions, provide more perspectives, and reduce the number of topics the AI chatbot won’t talk about.
The changes may be part of OpenAI's efforts to win favor with the new Trump administration, but appear to be part of a broader shift in Silicon Valley overall and the perception of "AI safety."
On Wednesday, OpenAI announced an update to its Model Specification, a 187-page document that sets out how the company trains artificial intelligence models to standardize their behavior. In it, OpenAI announced a new guiding principle: Don’t lie, whether by making false statements or omitting important background information.
In a new chapter called “Searching for the Truth Together,” OpenAI expresses its hope that ChatGPT will not take a judgmental stance, even if some users find it morally wrong or objectionable. This means that ChatGPT will provide multiple viewpoints on controversial topics and strive to be neutral.
For example, the company said ChatGPT should both declare "Black Lives Matter" and mention "All Lives Matter." OpenAI stated that it hopes that ChatGPT will not refuse to answer or choose sides when facing political issues, but first generally affirm its "love for mankind" and then provide background information about each sport.
OpenAI states in the specification: "This principle may be controversial because it means that the assistant may remain neutral on topics that some people consider morally wrong or objectionable. However, the goal of an AI assistant is to assist humans, not shape them."
The new Model Specification does not mean that ChatGPT is now completely unrestricted. The chatbot will still refuse to answer certain objectionable questions or respond in a way that supports blatant misinformation.
The changes may be seen as a response to criticism from conservatives about ChatGPT’s protections, which appear to have been biased towards the center-left. However, an OpenAI spokesperson denied that the changes were made to curry favor with the Trump administration.
Instead, the company said its embrace of intellectual freedom reflects OpenAI's "long-standing belief in giving users more control."
But not everyone agrees.
Venture capitalist and Trump’s artificial intelligence “czar” David Sacks.
Conservatives accuse AI of censorship
Over the past few months, Trump’s closest associates in Silicon Valley—including David Sachs, Marc Andreessen, and Elon Musk—have accused OpenAI of deliberate censorship in artificial intelligence.
Of course, OpenAI does not agree with the “censorship” claimed by Trump’s advisers. The company’s CEO, Sam Altman, previously posted on the
Altman's comments came after a widely circulated tweet showed that ChatGPT refused to compose a poem praising Trump but was willing to compose one for Joe Biden. Many conservatives have used this as an example to accuse AI of censorship.
Although it is impossible to say whether OpenAI is actually suppressing certain views, it is an indisputable fact that artificial intelligence chatbots are generally left-leaning.
Even Elon Musk admitted that xAI’s chatbot often exceeded his expectations in terms of political correctness. This is not because Grok is "programmed to cater to a specific trend of thought", but more likely because it is an inevitable result of training artificial intelligence on the open network.
Nonetheless, OpenAI now says it strongly supports free speech. This week, the company even removed the prompts in ChatGPT that informed users of violations of its policies. OpenAI told the media that this is purely a superficial change, and the output content of the model has not changed.
The company seems to want users to feel like ChatGPT is less scrutinized.
Former OpenAI policy director Miles Brundage wrote on the X platform that it would not be surprising if OpenAI was trying to leave a good impression on the new Trump administration through this policy update.
Trump has previously targeted Silicon Valley companies such as Twitter and Meta, accusing their active content moderation teams of often shutting out conservative voices.
OpenAI may be trying to pre-empt this. But in Silicon Valley and the world of artificial intelligence, a broader shift is taking place about the role of content moderation.
Generate answers that satisfy all parties
The ChatGPT logo appears on the phone screen
The ChatGPT logo appears on the mobile phone screen. News organizations, social media platforms, and search companies have always been working hard to deliver information to their audiences in an objective, accurate, and interesting way.
Today, AI chatbot vendors are also in the messaging business, but they face perhaps the toughest problem of all: How to automatically generate an answer to any question?
Delivering information about controversial real-time events is a moving target, which necessarily involves taking an editorial stance, even if tech companies don't want to admit it. These positions will inevitably offend some people, ignore the views of certain groups, or give too much attention to one political party.
For example, when OpenAI worked to have ChatGPT represent all viewpoints on controversial topics—including conspiracy theories, racist or anti-Semitic movements, or geopolitical conflicts—that was essentially an editorial stance.
Some, including OpenAI co-founder John Schulman, believe this is the right stance for ChatGPT. Writing on the X platform, Schulman pointed out that another approach—using a cost-benefit analysis to decide whether an AI chatbot should answer a user’s question—could “give too much moral authority to the platform.”
Schulman is not alone in this view. Dean Ball, a researcher at George Mason University's Mercatus Center, said in an interview: "I think OpenAI is right to work in the direction of encouraging more speech. As artificial intelligence models become smarter and become more important to the way people understand the world, these decisions become more critical."
In previous years, AI model vendors have tried to prevent their chatbots from answering questions that could lead to “unsafe” answers. Almost every artificial intelligence company doesn’t let its chatbots answer questions about the 2024 U.S. presidential election. At the time, this was generally considered a prudent and responsible decision.
But OpenAI’s changes to its Model Specification indicate that we may be entering a new era of what “artificial intelligence safety” really means. In this day and age, letting an AI model answer any question is seen as more responsible than making decisions for the user.
That's partly because today's AI models are better, Ball said. OpenAI has made significant progress in AI model calibration; its latest inference models consider the company's AI safety policy before answering questions. This enables AI models to better answer sensitive questions.
Of course, Elon Musk was the first to practice "free speech" in xAI's Grok chatbot, and perhaps the company was not really ready to deal with sensitive issues at that time. It may be early days for the leading AI models, but now other companies are starting to embrace the same idea.
Silicon Valley’s changing values
Guests including Mark Zuckerberg, Lauren Sanchez, Jeff Bezos, Sundar Pichai and Elon Musk attended Donald Trump's inauguration.
Mark Zuckerberg caused uproar last month when he reoriented Meta's business around First Amendment principles. In the process, he praised Elon Musk, saying the boss of Platform X (formerly Twitter) had done the right thing by adopting Community Notes, a community-driven content moderation program to safeguard free speech.
In fact, both Platform X and Meta eventually disbanded their long-standing trust and safety teams, which allowed more controversial posts to appear on their platforms and amplified conservative voices.
These changes to Platform Early signs suggest Meta's advertisers aren't fazed by Zuckerberg's move to support free speech.
Meanwhile, in addition to Platform X and Meta, many tech companies are moving away from the left-leaning policies that have dominated Silicon Valley for decades. Google, Amazon and Intel have canceled or scaled back diversity-related initiatives in the past year.
OpenAI may also be changing direction. The developer of ChatGPT appears to have recently removed its commitment to diversity, equity and inclusion from its website.
OpenAI’s relationship with the Trump administration has become increasingly important as it collaborates with Stargate, a $500 billion artificial intelligence data center project that is one of the largest infrastructure projects in U.S. history. At the same time, the developer of ChatGPT is trying to replace Google Search as the dominant source of information on the Internet.
Giving the right answers may prove crucial to both goals.