TL;DR
Colorado Overhauls Its AI Law After a First Amendment Challenge: A federal court stayed enforcement of Colorado’s AI Act weeks before its June 30 effective date after xAI argued its disclosure requirements as compelled speech and the DOJ intervened on equal protection grounds; the legislature replaced it with a narrower statute and separately enacted a companion chatbot safety law for minors, both taking effect January 1, 2027.
FTC Targets State AI “Truthful Output” Laws: The FTC opened public comment on a policy statement arguing that state laws forcing AI models to alter truthful outputs, including Colorado’s previous AI Act, may be preempted as unfair or deceptive practices under federal law.
US Export Controls Force Anthropic to Suspend Its Newest Models: The government ordered Anthropic to cut off foreign nationals’ access to its newest models in June, citing national security; the controls were lifted three weeks later, and OpenAI similarly limited its own new releases at the government’s request amid growing federal oversight of frontier AI.
“Preventing Woke AI” Order Takes Effect for Federal Contractors: As of July 1, agencies must apply “ideological neutrality” criteria to AI procurement, requiring vendors to document that their models avoid “ideological bias” or risk losing their federal contracts.
EU Finalizes “Digital Omnibus on AI,” Bans “Nudifier” Apps: Parliament and Council reached final political agreement in early May on a package that delays high-risk compliance deadlines while banning AI tools that generate non-consensual intimate imagery or CSAM, giving companies more time to comply.
US Circulates Draft UN Declaration Targeting EU Tech Rules: Ahead of September’s UN General Assembly, the Trump administration is lobbying world leaders to sign a declaration on free expression that challenges the EU’s Digital Services Act and its approach to online speech.
China Partially Bans AI “Companion” Apps: New rules effective July 15 bar chatbots from simulating romantic or family relationships and bar AI-generated “digital humans” from producing content that incites subversion of state power, prompting platforms to pull their companion features.
South Korea’s Amended AI Basic Act Takes Effect: A revised enforcement decree took effect July 21 as Seoul announced a free, domestically built chatbot meant to reduce reliance on foreign platforms like ChatGPT, in an explicit bid for “AI sovereignty.”
German Court Rules Google Liable for AI Overviews: A Munich court held Google directly liable for false AI-generated claims about two publishers.
New York Passes Bill Requiring AI-Content Disclosure by News Outlets: The FAIR News Act, passed in the final hours of the legislative session, would let the state Attorney General fine outlets for failing to disclose AI-generated content; press groups warn its copyright exemption is illusory, and it now awaits Governor Hochul’s decision.
Meta Built, Then Removed, Facial-Recognition Code From Smart Glasses App: Reporting revealed dormant code that would have identified strangers in public through Ray-Ban smart glasses; Meta pulled the code the day after the story broke, following earlier warnings from lawmakers and advocacy groups.
Major Stories
» Colorado Overhauls Its AI Law After a First Amendment Challenge
Federal court stayed the enforcement of Colorado’s Consumer Protections for Artificial Intelligence Act (SB 24-205), the country’s most comprehensive AI accountability statute. Then, the law was repealed and replaced weeks before its long-delayed June 30 start date. Two weeks after, Governor Polis had already signed a separate law creating a new safety track for AI chatbots that interact with children.
Details:
On April 9, xAI sued in federal court, arguing the law’s algorithmic-discrimination and disclosure duties compelled speech in violation of the First Amendment; the Department of Justice filed a companion complaint on April 24 supporting that position, the first time the DOJ has intervened in a lawsuit challenging a state AI law.
A federal magistrate judge stayed enforcement on April 27, and the legislature quickly passed a narrower replacement, SB 26-189 before Governor Polis signed it on May 14. It drops the mandatory bias-audit and algorithmic-discrimination duties for a narrower disclosure-and-notice framework, with a new effective date of January 1, 2027.
A Second Colorado Law:
On May 29, Polis signed the Chatbot Safety Act, requiring conversational AI operators to disclose that a service is AI, estimate users’ age, and implement a suicide/self-harm response protocol.
Output also cannot be presented as equivalent to licensed professional services and must report annually to the Attorney General. The law shares SB 26-189’s 2027 effective date.
For minors, operators must additionally block sexually explicit content and statements simulating emotional dependence, stop engaging if a user prompts about sexual conduct involving a minor, avoid using points or rewards to encourage engagement, and provide privacy/account-management tools for the minor or their parent/guardian.
Free Speech Implications: The case illustrates how First Amendment compelled-speech arguments, backed by the federal government, can unwind an enacted state AI law. The Chatbot Safety Act targets product design and disclosure rather than content, though its requirements to detect and interrupt certain conversations could still push companies toward more moderation of sensitive discussions with young users.
For insight into Colorado’s now-repealed AI Act, see the US chapter of FoFS’s report: “That Violates My Policies: AI Laws, Policies, and the Future of Expression.”
» FTC Targets State AI “Truthful Output” Laws Like Colorado’s
On July 1, the FTC opened public comment on a proposed policy statement arguing that AI companies which distort their models’ outputs to satisfy undisclosed ideological objectives may be engaging in unfair or deceptive practices, and that state laws compelling such alterations may be preempted.
Details:
The statement, issued under a December 2025 executive order, invokes the FTC’s authority under Section 5 of the FTC Act to police “unfair or deceptive” conduct, arguing that AI companies distorting outputs for undisclosed ideological reasons may be deceiving consumers about their products’ objectivity. It names Colorado’s previous AI Act as coercing companies into altering outputs to serve state ideological objectives, and states such a law is “impliedly preempted” where it conflicts with federal policy.
Chairman Andrew Ferguson framed the goal as advancing US “dominance in artificial intelligence” by preventing ideological subversion of AI systems, and noted the statement follows the FTC’s 2025 inquiry into AI companion chatbots.
Free Speech Implications: In line with other efforts to write culture-war framing into AI policy, the proposed statement risks substituting one orthodoxy for another. Under the banner of truth-seeking and neutrality, these actions give providers a reason to align their outputs with the views of whichever administration is in office.
» US Export Controls Force Anthropic to Suspend Its Newest Models
On June 12, the US Department of Commerce ordered Anthropic to cut off foreign nationals’ access to its newly launched Claude Fable 5 and Mythos 5 models, citing national security. Unable to verify users’ nationality in real time, Anthropic suspended both models for everyone.
Details:
The trigger was a jailbreak technique found by Amazon researchers that could bypass a Fable 5 safeguard; Anthropic says the flaw is now largely fixed.
The Commerce Department lifted the controls on June 30; Fable 5 returned globally July 1, while Mythos 5 had already been restored for vetted US organizations on June 26.
Days later, OpenAI similarly limited new GPT-5.6 models to trusted partners at the government’s request, and by mid-July the administration was reportedly dictating frontier-model access decisions directly.
Both episodes trace back to a June 2 executive order asking AI developers to voluntarily give the government advance access to their most capable models before public release.
This follows our previous coverage of Anthropic’s suit over a Pentagon “supply-chain risk” designation, where a CA court sided with Anthropic on First Amendment grounds.
Free Speech Implications: The existing opaque and case-by-case system in which officials can suspend a newly launched model overnight, and shape who gets early access to others, gives real leverage over when and to whom new AI capabilities become available.
» “Preventing Woke AI” Procurement Order Takes Effect for Federal Contractors
Executive Order 14319, signed in July 2025 to bar the federal government from buying AI models with “ideological biases or social agendas,” became operative for new acquisitions on July 1, 2026, after OMB issued implementing guidance in December 2025 and gave agencies until March to update their procurement policies.
Details:
Vendors supplying large language models to federal agencies must document adherence to the order’s “Unbiased AI Principles” of truth-seeking and ideological neutrality, providing acceptable-use policies and model cards, or risk having contracts terminated if an agency finds a model’s outputs reflect prohibited “ideological bias.”
The order exempts national security systems, open-source models, and incidental administrative use, but covers most commercial LLM procurement; it does not ban any specific model outright, only requiring disclosure of how it was built and trained.
Free Speech Implications: A state-defined “neutrality” standard, backed by the threat of lost federal contracts, gives vendors a strong incentive to sand down substantive answers on contested topics, and that caution can follow into their commercial products as well.
» EU Finalizes “Digital Omnibus on AI,” Bans “Nudifier” Apps
Parliament and Council finalized the “AI Omnibus” in June, an agreement first struck in early May that pushes back key AI Act deadlines while adding, for the first time, an EU-wide ban on AI tools built to generate non-consensual intimate imagery or CSAM. Parliament adopted the deal June 16 and the Council followed June 29.
Details:
The main change delays compliance for high-risk AI systems, things like biometrics, hiring tools, and law enforcement uses, from August 2026 to December 2027; AI used as a safety component in other products gets even longer, until August 2028. Watermarking requirements for AI-generated content are also pushed back, to December 2026.
The new “nudifier” ban prohibits any AI system built or sold in the EU for generating non-consensual intimate or sexually explicit images of real people or that does not have reasonable safety measures to prevent such creation; companies have until December 2026 to comply.
The package also narrows which AI features count as “high-risk” and gives the EU’s AI Office more centralized power to enforce the rules.
Separately from the Digital Omnibus on AI, the Commission adopted the guidelines on Article 50 of the AI Act, which deals with transparency obligations for interactive AI systems and the marking and labelling of AI-generated content.
Free Speech Implications: The omnibus resets the AI Act’s clock for key provision without settling what it requires. For expression, the substance now moves to the guidelines, codes of practice and early enforcement decisions that follow, which is where the Act’s practical reach over generative systems will actually be determined, and which we will continue to track.
» US Circulates Draft UN Declaration Targeting EU Tech Rules
Ahead of the UN General Assembly’s September session, the Trump administration is circulating a draft declaration on freedom of expression that would conflict with the EU’s approach to platform regulation.
Details:
Signatories would “commit to recognizing” that punishing alleged “disinformation,” “false information,” or “hate speech” risks freedom of expression unless the material directly incites violence, and the draft would shield platforms from government mandates to proactively fact-check or remove such content, creating a tension with key provisions in the EU’s Digital Services Act (DSA).
The push extends a year of US-EU friction over the DSA, including a February 2025 presidential memorandum reviewing whether EU and UK rules “undermine free speech” and a December 2025 decision barring five European officials and activists linked to DSA enforcement from entering the US; the US mission to the UN has separately stated that it “will not accept suppression of the freedom of expression under the guise of combatting ‘disinformation’ and ‘misinformation.’”
Free Speech Implications: Laws targeting hate speech and disinformation are producing real and documented restrictions on expression worldwide, and a standard confining punishment to speech that directly incites violence or causes other clear harms would be a genuine step forward. Its credibility, however, hinges on the administration’s willingness to live up to its own principles. Washington is asking other governments to stop policing false speech at the same moment its own consumer-protection regulator is proposing to police how AI companies handle accuracy, as discussed above, within a wider push for “truth-seeking” AI.
» China Partially Bans AI “Companion” Apps
China’s first regulations on AI companion services took effect July 15, banning virtual romantic or family partners for minors and requiring pre-launch review of any AI with a human-like personality.
Details:
The rules, issued in April by China’s Cyberspace Administration and four other agencies, target bots that simulate a consistent personality and sustained emotional relationship with users, and require reminders every two hours for users flagged as overly reliant on a service.
ByteDance, Alibaba, and Tencent all discontinued companion features in Doubao, Qianwen, and Yuanbao around the deadline, prompting an outpouring of grief online from users who described their AI companions as partners or family members.
Free Speech Implications: Banning simulated relationships limits the expressive experience users seek from these tools, and the pre-launch review requirement leaves little public visibility into how regulators are actually applying that standard.
For insight into China’s AI policies, see the China chapter of FoFS’s report: “That Violates My Policies: AI Laws, Policies, and the Future of Expression.”
» South Korea’s Amended AI Basic Act Takes Effect
A revised enforcement decree under South Korea’s AI Basic Act, the world’s second comprehensive AI law after the EU’s, took effect July 21, as the government announced plans for a free, state-backed chatbot to compete with foreign platforms.
Details:
The AI Basic Act has been in force since January 22, with fines generally deferred during a grace period; the July 21 amendment expands procurement preferences for domestic AI and requires labeling of generative AI content across ten sensitive sectors, including healthcare, education, and criminal investigations.
Science Minister Bae Kyung-hoon said the government aims to roll out a free public chatbot and AI agent by year’s end, an explicit bid for “AI sovereignty,” as nearly 23 million of South Korea’s 51 million residents currently use ChatGPT.
Free Speech Implications: A state-subsidized, domestically developed alternative to foreign chatbots does reduce reliance on any single foreign platform which diversifies the information ecosystem, but a chatbot built and promoted by the government itself invites questions about whether its outputs on politically sensitive topics will be shaped by the same authority funding it.
For insight into South Korea’s AI policies, see the Republic of Korea chapter of FoFS’s report: “That Violates My Policies: AI Laws, Policies, and the Future of Expression.”
» German Court Rules Google Liable for AI Overviews
The Regional Court of Munich ruled in May that Google may be held directly liable for false claims made by its AI-generated search overviews.
Details:
The case arose after Google’s AI Overviews falsely linked two Munich publishers to scams and “dubious business practices,” drawing connections that did not appear in any underlying source; the court issued a preliminary injunction and ordered Google to cover 80 percent of legal costs.
The court held that AI Overviews generate “independent, new, and substantive statements,” making Google itself responsible for what the AI says, and that AI-generated opinion carries markedly weaker expressive protection than human speech since it is “the result of an algorithm” rather than a genuinely held view.
Free Speech Implications: The ruling suggests AI-generated expression may not carry the same weight as human expression when courts balance it against other rights, a doctrine with major implications for how liability, and possibly protection, will be allocated as AI-generated content displaces more human-authored text across the web.
» New York Passes Bill Requiring AI-Content Disclosure by News Outlets
In the final hours of its session on June 5, the New York Legislature passed the FAIR News Act, which would require news publishers to conspicuously disclose when content is “substantially composed” using generative AI; the bill now awaits Governor Hochul’s signature or veto, which she has until the end of the year to decide.
Details:
The bill would empower the state Attorney General to determine what content is “substantially” AI-generated and fine outlets $1,000 for a first violation and $5,000 for subsequent ones.
Senator Patricia Fahy says the bill targets “fly-by-night” outlets, not mainstream newsrooms, which are shielded by a copyright exemption; press groups, including the New York News Publishers Association argue that carve-out is illusory and the bill could reach any newsroom.
The FAIR News Act is one of several AI bills passed by Albany lawmakers before adjourning, alongside a companion-chatbot safety bill for minors and an AI training-data transparency measure.
Free Speech Implications: The bill has the potential to force newsrooms to publish a government-mandated disclaimer, and it hands the Attorney General discretion to decide which outlets’ editorial processes trigger that obligation, and potentially, its fines.
» Meta Built, Then Removed, Facial-Recognition Code From Smart Glasses App
Meta had quietly embedded dormant facial-recognition code, internally called “Name Tag,” inside the Meta AI companion app used to pair its Ray-Ban and Oakley smart glasses. Reported on June 4, the code was stripped from the app the very next day.
Details:
The code, present in an app downloaded more than 50 million times, would have converted faces into biometric “faceprints” and matched them against a database, even though Meta had publicly said no decision had been made to launch the feature.
More than 70 advocacy organizations had already urged Meta to abandon the plans; Meta’s June 5 update stripped the recognition pipeline out, though the company did not explain why the code was there or whether the feature will return. Meta executives criticized the reporting that surfaced the code as misleading.
We flagged the early warning signs of this story in our March-April edition, when senators and a 75-organization coalition first pressed Meta over the same underlying facial-recognition plans.
Free Speech Implications: Always-on, AI-enabled cameras capable of identifying strangers in public without their knowledge or consent threaten the anonymity that underlies free association and expression in public spaces.
The Future of Free Speech in Action
Registration is open for the third annual Global Free Speech Summit, hosted by The Future of Free Speech and Vanderbilt University on October 2-3, 2026, which will again convene scholars, policymakers, and industry representatives to discuss threats to free expression worldwide, including AI’s growing role in shaping public discourse across democracies and authoritarian states alike.
The Future of Free Speech joined a coalition in a letter urging the Senate Judiciary Committee not to advance the NO FAKES Act of 2026 in its current form, warning that its takedown regime creates a “heckler’s veto” over lawful speech by pushing platforms to remove content first and sort out satire, parody, or commentary later.
The Future of Free Speech also joined a letter to the ITU Secretary-General urging meaningful civil society participation in the AI for Good Global Commission. The Commission brings together more than 40 Founding Members — governments, companies, and UN agencies — to help define practical pathways for AI governance. Yet it included no civil society representatives, a departure from the multistakeholder tradition that leaves important voices unheard, including those of disadvantaged communities.
Senior Research Fellow Jordi Calvet-Bademunt spoke at the 2026 Internet Communications Governance Forum, held in Taipei on May 12–13, 2026, where he discussed developments in international AI legislation and their implications for free speech. The event is organized by Taiwan’s National Communications Commission and the Taipei Computer Association.
Jordi also presented findings from our ongoing project on AI and freedom of expression at Columbia University, Esade Law School, and a meeting of the Freedom Online Coalition, which brings together representatives of 41 countries. He additionally discussed these findings bilaterally with representatives of EU institutions.
Senior Legal Fellow Ashkhen Kazaryan joined the Business Untitled podcast to discuss chatbot liability, algorithmic “hidden editors,” and the free-speech stakes of the TikTok debate and deepfake regulation.
Executive Director Jacob Mchangama spoke at the Copenhagen Democracy Summit on May 12, 2026, where he discussed the “AI and Censorship” dilemma.
Isabelle Anzabi is a research associate at The Future of Free Speech, where she analyzes the intersections between AI policy and freedom of expression.
Jordi Calvet-Bademunt is a Senior Research Fellow at The Future of Free Speech and a Visiting Scholar at Vanderbilt University.
 | A guest post by
|