The same "trust and safety" teams that spent years censoring political speech on social media are now running AI safety departments — and if you think that's a coincidence, Allum Bokhari told The Ezra Levant Show it should come as no surprise.

The managing director of the Foundation for Freedom Online and author of Deleted: Big Tech's Battle to Erase the Trump Movement joined Ezra Levant for an  in-depth discussion on the topic on Wednesday's episode.

Bokhari laid out what he calls machine learning fairness — a field that blends critical race theory with computer science. Under this framework, AI safety doesn't mean preventing bioweapons or cyberattacks. It means ensuring AI doesn't produce outputs that notice racial or gender differences in ways progressives find inconvenient.

Some Democrat-led states in the U.S. have already moved toward laws on what they call "algorithmic discrimination" — a phrase used in at least one Colorado AI bill that, Bokhari noted, effectively permits algorithmic discrimination only when it is used to redress "historical imbalances."

"What they actually mean," he said, "is it's allowed, but only if it's discriminating against white males."

Go Ad Free

In contrast, the Trump administration's AI policy takes a narrower view, focusing safety definitions on genuine national security and cybersecurity concerns rather than importing progressive ideology into AI guardrails. 

The existence of Grok, Elon Musk's AI model, matters precisely because it is explicitly designed to be truth-maximizing rather than politically constrained. "It's important that there's not just one AI company," Bokhari explained, "so that it doesn't get controlled from the top down by very ideological forces."

The discussion then shifted to another major player in the AI field, Anthropic.

Bokhari described the company as a "black sheep" in the industry, the only major player actively lobbying against open-weight models. Open-weight AI means the underlying algorithm is accessible to anyone, allowing people to run AI on their own devices without depending on a single company.

Every other significant AI company supports this direction, while Anthropic fights against it. "This will determine whether these technologies will be open to all," Bokhari said, "or whether it'll just be a small group of tech elites, perhaps working hand in glove with nation states, that control this."

Ezra noted that Anthropic's CEO Dario Amodei was recently quoted predicting that Anthropic would eventually be the only AI company in the world — and did not deny the quote when pressed. "That's straight out of a dystopian science fiction movie," Ezra said. Bokhari agreed. "When you hear people like Amodei saying we're going to be the only AI company, those fears no longer seem like conspiracy theories. They seem entirely valid."

For those worried about AI displacing jobs, Bokhari said the most important question is who controls the technology.

If AI development becomes monopolized by one ideologically captured company or a handful of nation-state-aligned players, the risks are far greater than job displacement. He compared unilaterally restricting American AI development — as the anti-data-centre movement effectively advocates — to the U.S. refusing to build nuclear weapons in the 1950s and leaving the field entirely to the Soviets.

"If AI is really dangerous," he said, "then you at least need America's adversaries to agree to restrict them before you unilaterally do anything."

Ezra also played a brief clip from his conversation with David Sacks, Trump's AI and crypto adviser, at the World Economic Forum, in which Sacks acknowledged the growing divergence between American and European approaches to free speech online — noting that the State Department had sanctioned European NGOs actively promoting internet censorship, and that 12,000 people per year in the UK are arrested for speech crimes.

"How is freedom of speech going down in Europe," Sacks said, "and going up in the Middle East? Think about that trend."