Digital Freedom, AI Regulation, and the Fight for the Western Internet | The a16z Show
Summary
Sarah Rogers agrees with Tyler Cowen’s phrase “AI with a Western soul” and argues that a Western AI stack is a national-security and soft-power priority. AI will underpin much global communication and commerce, so systems that reason in an individualistic, rules-based way and prioritize user consent could advance freedom.
Among the major regulatory risks are rules that undermine fair-use protections or make model development legally unpredictable. Rogers flags foreign copyright regimes, disclosures that could expose model weights, vague assessments of hate speech, civil discourse, and well-being, and draft laws imposing strict criminal liability when an LLM is merely capable of generating certain content—even content that might not be visible under the First Amendment.
European laws and ostensibly content-neutral enforcement can reach American platforms, users, and revenue. Rogers’s central example is Thierry Breton’s August 2024 warning before Elon Musk’s Trump interview, which she says linked airing the interview to a separate X investigation that later culminated in a €120 million fine; other European laws can expose companies to penalties reaching 6% of global revenue.
The State Department’s digital-freedom posture has shifted from upstream information control toward user agency. Rogers says adversary information operations are real but prior efforts went overboard. Her office continues work against malware, spyware, and cyberattacks while favoring content provenance, censorship-circumvention VPNs, and Community Notes over opaque government or NGO choke points.
For AI in national defense, Rogers says consequential policy questions should be handled through courts and democratic deliberation rather than executive or employee fiat. Questions about autonomous weapons, surveillance, and data synthesis should be governed by the rule of law.
Her practical prescription for founders is crisp regulation plus viewpoint neutrality—not zero moderation. Government should avoid arbitrary regulatory cudgels while protecting companies from foreign coercion; platforms can still let users filter spam, pornography, or foreign-provenance content because those are not viewpoint-based distinctions.
Deep dive
1. Public diplomacy is being rebuilt around user agency
Rogers defines public diplomacy as the American government’s relationship with foreign publics, spanning Fulbright educational and cultural exchanges, fast media-response assets, and engagement with the information environment—the “operating system” on which international conversations run.
She says the prior administration’s Global Engagement Center pursued censorship efforts, including contacting platforms about alleged disinformation in Charlie Kirk tweets; that apparatus was housed in her part of the State Department and was the subject of the Murthy v. Missouri litigation. After the reorganization, she acquired the Digital Freedom Office and new internet-freedom authorities under the National Defense Authorization Act.
Her account of internet policy begins with Arab Spring and Occupy Wall Street-era enthusiasm for disintermediated conversation, which challenged entrenched institutions. After the Arab Spring, she thinks officials grew nervous that there could be an “American Spring”; Trump and Brexit in 2016 then intensified efforts toward “disinformation” curation and suppression of “adverse narratives.” The intentions were often prosocial and adversary information operations are real, she concedes, but officials “went overboard.”
Rogers says her office is pursuing “transparency, truth and reconciliation” regarding prior censorship while retaining work against malware, spyware, and foreign cyberattacks. She supports content provenance so users can assess where content came from, whether it was AI-generated, and whether it is true. She rejects opaque government-funded intermediaries deciding which arguments—such as arguments about pediatric transgender medicine—users may see. She is friendlier to censorship-circumvention VPNs and X’s Community Notes because they place power with users or the crowd transparently.
2. A Western AI stack is strategic infrastructure
Rogers agrees with economist Tyler Cowen’s phrase “AI with a Western soul”: models that reason in an individualistic, rules-based way and prioritize user consent. Because AI could become the underlying reasoning model for much of global communication and commerce, she says Western-stack proliferation is a priority for the entire administration and for anyone who cares about freedom.
Project Maven is her double-edged precedent: it catalyzed awareness that technological innovation and national security were one interest, but also triggered ideologically driven employee revolts at Google. AI makes both dynamics more consequential.
Her printing-press and telegraph analogies frame today’s regulatory instinct: each communications revolution provokes fear about social effects and an urge to “put the innovation back in the bottle,” or let legacy institutions control it. Rogers argues that American values and interests instead favor using technology “to make people free.”
3. Europe’s speech enforcement creates cross-border platform risk
Rogers’s catalytic example is Thierry Breton’s August 2024 letter to Elon Musk before his interview on X with then-presidential candidate and former president Donald Trump. The interview had not happened, yet Breton threatened regulatory penalties if Musk aired it and referred to a separate ongoing investigation against X.
That investigation was ostensibly about whether blue checks were assigned authentically and whether X’s algorithm was transparent. Rogers says Breton’s letter effectively warned that allowing Trump to speak would increase X’s exposure to adverse findings in that investigation, which later culminated in a €120 million fine. She sees this as viewpoint-skewed enforcement of ostensibly content-neutral rules—“insidious and inevitable” when the subject is politically pitched.
The cross-border stakes rise when laws make it illegal to insult a politician, as Rogers says is the case in Germany, or when some UK prosecutors have recently argued for treating blasphemy against Islam as illegal. Transposed to a global internet, such laws can expose American companies to penalties reaching 6% of global revenue. Rogers contrasts this with Russia and China, which she says firewall off their networks rather than purporting to fine American companies for allowing Americans to engage in First Amendment-protected political speech.
Rogers describes confrontational conversations on X alongside constructive diplomatic discussions. She says the U.S. values Europe as a safe, strong, and prosperous ally with which it comprises “1 civilization” and shares interests, but the goal is not identical speech regimes; it is to prevent an “insidious and sweeping censorship contagion.”
4. Copyright and liability rules will shape AI development
Rogers calls Section 230 of the Communications Decency Act and the fair-use doctrine structural features of the internet that grew out of American law. Her analogy for AI training is a kindergarten class learning from a library of books: teaching the class would be fair use, and a lot of courts are saying that paying for the same books to train an LLM is fair use as well.
She worries that other countries, including perhaps the EU, will treat copyright differently in ways that could be devastating to AI development. Transparency is understandable, she says, but forcing companies to disclose information that lets foreign adversaries reverse-engineer model weights would compromise the American and Western competitive edge.
She also warns against applying Europe’s content rules for “very large online platforms” and “very large online search engines” to AI. The risk assessments those rules demand—covering hate speech, civil discourse, or harm to well-being—can be vague, especially because LLMs can emit unpredictable responses despite extensive safety architecture.
Rogers says she has seen draft legislation imposing strict criminal liability if an LLM is even capable of generating certain content—not just child pornography, but content that might not even be visible under the First Amendment. She argues that such a strict-liability regime would degrade Section 230’s protection layer and create adverse incentives against creative model training.
5. Rule of law—not corporate fiat—should govern AI alignment
On Department of War contracting, Rogers defers questions about which companies meet statutory thresholds to the department’s lawyers. Her broader national-security requirement is that AI retain its “Western soul.”
She rejects letting Silicon Valley executives or tech workers settle by fiat questions about autonomous weapons’ “kill shots,” surveillance, or data synthesis. She cites a prior Twitter policy under which users could not call a convicted sex offender male, arguing that such decisions should instead be addressed through courts, statehouse steps, and other “crucibles of democratic deliberation” that have served the country for 250 years.
Her domestic playbook is “crisp and principled” regulation that makes compliance clear and favors viewpoint neutrality. She points to the X blue-check investigation and debanking cases—including her representation of the NRA, where she says ostensibly viewpoint-neutral rules were disproportionately enforced against banks serving pro-gun groups—as warnings that neutral powers can become arbitrary regulatory cudgels.
Viewpoint neutrality still permits user-controlled filtering of spam, pornography, or content with foreign provenance. Those are not viewpoint-based distinctions, and Rogers says regulation should be more accommodating of that kind of moderation.
Government must also defend American platforms when foreign governments threaten their interests and political freedoms. Rogers’s analogy is that if Washington threatened French platforms Le Monde or Vivendi for hosting an interview with Emmanuel Macron, France would object; the U.S. should not accept the reverse situation either.