If AI companies can deny tools to lawful groups with disfavored views, they can tilt markets, speech, and politics without firing a shot.
Many of my articles about artificial intelligence focus on individual liberty and personal autonomy in the age of AI.
Conservatives worry about a future in which AI systems silently coerce individuals, institutions, or society itself in one direction or another. After Big Tech censorship, ESG mandates, and “Great Reset” schemes pushed by powerful institutions, those worries look warranted. To many people, however, they still feel theoretical.
Denying access to artificial intelligence tools on ideological grounds means denying people access to technologies that AI leaders insist will define the future economy.
Then I saw a real-world example of AI-era censorship. The implications are chilling.
Several years ago, conservative media began sounding the alarm over the power of social media companies such as Facebook and Twitter to stifle conservative speech. Stories regularly emerged about conservative commentators and organizations being shadow-banned, demonetized, deplatformed, or otherwise suppressed.
Digital media had become one of the most powerful ways to reach large audiences. Critics rightly warned that Big Tech companies wielded undue influence over public discourse. In many cases, they decided which ideas would be amplified, which would be buried, and which would be treated as too dangerous for polite society.
For years, I’ve warned that Americans could experience the same thing on a much larger scale if AI companies embed similar biases into their foundational models. We have already seen small precursor examples. But the example I recently witnessed blew my mind.
A colleague who works on website production at the Heartland Institute told me he had been working with Anthropic’s Claude to design a website for our organization. In a back-and-forth exchange, he asked Claude to help brainstorm ways to increase Heartland Institute branding on the new climate-conferences.com website.
Instead of helping with a simple design request, Claude refused. The request did not ask Claude to fabricate data. It did not ask Claude to impersonate anyone, threaten anyone, harass anyone, defraud anyone, or break the law. The request asked for website branding help.
Anthropic’s AI refused because the Heartland Institute has been labeled a “climate denial” group.
That’s a big deal with big-time implications.
To understand the implications, remember what the AI industry has spent the past several years telling Americans: Artificial intelligence will transform everything. It will supercharge productivity, accelerate scientific discovery, revolutionize education, reshape medicine, streamline business operations, and push nearly every industry into a new era of growth. Analysts estimate that generative AI could add trillions of dollars in value to the global economy. Anthropic’s own research on Claude has found that AI is already being used across a wide range of occupations, especially in software development and writing.
So what happens when the same companies building this supposedly essential economic infrastructure decide that certain people, organizations, industries, or ideas do not deserve access to it?
At best, disfavored users fall behind. At worst, they fail completely.
If one organization can use AI to design websites, polish messaging, automate research, improve outreach, generate graphics, write code, and sharpen communications, while another is blocked because its mission is politically disfavored, the playing field no longer exists. The AI company has tilted the marketplace.
Today, the target is the Heartland Institute. Tomorrow, it could be a pro-life organization, a religious school, a gun rights group, a parents’ organization, a dissident scientist, or any business that offends whatever ideological framework has been embedded inside the model.
That is the deeper danger revealed by Claude’s refusal: AI tools have grown beyond mere chatbots. They are becoming assistants, designers, researchers, coders, marketers, editors, analysts, and strategic consultants.
Denying access to those tools on ideological grounds means denying people access to technologies that AI leaders insist will define the future economy.
In that world, AI companies choose winners and losers.
Imagine if in the early age of electricity, power companies decided which businesses, churches, newspapers, schools, or civic organizations were allowed to plug into the grid based on politics. Imagine if a power company told a disfavored newspaper it could not use electricity to print more persuasive editorials, or told a religious school it could not use electricity to expand its reach because its beliefs were considered harmful.
Americans would have immediately recognized the danger.
AI companies are now doing something similar under values most Americans never voted for, rules most users never accepted, and ideological judgments no private company should be trusted to apply at civilization-altering scale.
The most troubling phrase in Claude’s response was “downstream effect.” That phrase reveals the logic behind the refusal. Claude judged the potential impact of helping Heartland communicate more effectively. It refused design assistance because making the website more polished and persuasive might help the organization’s message reach more people.
That should concern everyone, regardless of his views on climate change.
Today, the target is the Heartland Institute. Tomorrow, it could be a pro-life organization, a religious school, a gun rights group, a parents’ organization, a dissident scientist, or any business that offends whatever ideological framework has been embedded inside the model.
