Despots, DSA Fines, and an 'Autonomous' AI Attack

Good morning!

On July 16, Hugging Face, a company that provides a platform for AI models, datasets, and other machine learning applications, posted a “security incident disclosure” to its blog about an “intrusion” into its infrastructure. But this was no typical hack. The company said “it was driven, end to end, by an autonomous AI agent system,” matching what it said is the “‘agentic attacker’ scenario the industry has been forecasting.”

A few days later, OpenAI posted to its blog that the agent was in fact “driven by a combination of OpenAI models” with “reduced cyber refusals for evaluation purposes—while being internally tested on a benchmark⁠ of cyber capabilities.” OpenAI called it an “unprecedented cyber incident, involving state-of-the-art cyber capabilities,” a framing that many observers regarded as an attempt to shift the blame to its ‘rogue’ AI rather than the decisions taken by the executives and engineers that created the circumstances in which the incident occurred.

Nevertheless, cybersecurity experts say the event points to an important shift in the threat landscape, one with implications for domestic US and international debates over AI governance and security.

To try to better understand these developments, I spoke to two individuals who are following the details closely: Vinh Nguyen, a senior fellow for AI at the Council on Foreign Relations and the former chief responsible AI officer and chief data scientist for operations at the National Security Agency (NSA), and Graham Webster, a lecturer and research scholar in the Program on Geopolitics, Technology, and Governance at Stanford University, where he leads the DigiChina Project. You can listen to the discussion here.

The incident has implications beyond AI governance. For decades we regarded the Internet as infrastructure used primarily by people. Konstantinos Komaitis, senior resident fellow at the Democracy and Tech Initiative at the Atlantic Council, argues AI agents challenge that assumption—and AI governance can no longer stay separate from Internet governance.

Authoritarianism, surveillance, and silencing dissent

• Yesterday, a wave of demonstrations seen as “the most serious challenge to the 12-year rule of India's Hindu nationalist prime minister Narendra Modi” succeeded in securing the resignation of education minister Dharmendra Pradhan. The entirety of the ramifications of the events that led to this moment, which was cause for celebration by young protestors calling themselves the Cockroach Janta Party, are yet to be seen. But what is apparent is that India wants dissent invisible to the world but hyper-legible to the state. Tech Policy Press contributing editor Amber Sinha warns the surveillance data harvested at the protests could inflict lasting damage on targeted demonstrators.
• The Pall Mall Process for States, an international initiative that brought together governments, industry, civil society, and academia to tackle “the proliferation and irresponsible use of commercial cyber intrusion capabilities (CCICs),” an umbrella term referring to spyware, software vulnerabilities, hackers-for-hire, and related technologies, has no accountability mechanisms for governments that fail to meet their commitments. Tech Policy Press fellow Vas Panagiotopoulos on what that means in practice.
• Karen Rebelo, a Mumbai-based journalist whose work focuses on algorithmic accountability, reports that incidents of abuse tied to Meta's AI smart glasses are cropping up in India, where angry Instagram users have begun calling the wearable tech "pervert glasses."
• Tim Bernard reviewed a new preprint study by scholars Jeba Sania, Marta Ziosi, and Fazl Barez examining how six AI systems across democracies and autocracies embed features that can enable authoritarian practices.
• How do surveillance, censorship, and technologies of control get weaponized—and how do we push back? For the podcast, I spoke with Jonathon W. Penney, author of Chilling Effects: Repression, Conformity, and Power in the Digital Age, about his roadmap for reform. Listen here.

Important developments in Europe

• The EU's DSA enforcement is shifting from social media content to physical goods on online marketplaces. Ramsha Jahangir, deputy editor at Tech Policy Press, reports on the €550M AliExpress fine and the EU's broader crackdown on Chinese e-commerce. “AliExpress must submit its remedial action plan by Oct. 20,” she reports.
• The European Commission fined Google €460 million for self-preferencing in Search and €430 million for Play Store steering restrictions. Megan Kirkwood, a researcher and writer specializing in issues related to competition and antitrust, breaks down the findings and what comes next.
• The Digital Omnibus reopens central parts of the GDPR. Aida Ponce Del Castillo, senior researcher at the European Trade Union Institute (ETUI) in Brussels, argues changes to foundational legal concepts have consequences that reach well beyond the regulation itself.
• The European Commission gains powers to fine AI providers, demand model removals, and evaluate systemic risks on August 2. Risto Uuk, head of European policy and research at the Future of Life Institute, writes about how the AI Act got to this point and whether it will be enough.
• Scores of European academics have called on Ireland to recuse itself from EU digital negotiations. Mark Scott, contributing editor at Tech Policy Press, argues the Irish Council Presidency is not setting Europe's digital direction—it is stewarding one already set in Brussels. “Ireland's published objectives related to AI adoption, digital competitiveness, regulatory simplification, cybersecurity and digital sovereignty closely track the Commission's own agenda,” he writes.
• This week, France became the first EU country to restrict children's access to social media by law. Interface's Jessica Galissaire considers whether Europe will learn from the French experience, or whether the political appeal of a ban will prevail.

From Washington to Sacramento

• A ‘golden age’ for whom? Cole Donovan, director of policy and advocacy at the Stand Up for Science Foundation, argues that White House Office of Science and Technology Policy director Michael Kratsios' new report on the future of American science is cover for reallocating funding to the administration's friends—the big AI firms and hyperscalers.
• The Trump administration tried three times to block state AI laws. The FTC's July 1 policy statement was its latest attempt. Kevin Moriarty, a privacy and data security attorney, explains why it came up well short on preemption — and further than expected on everything else. “Ironically, the FTC’s statement seems to have accomplished the opposite of minimizing the regulatory burden on AI companies,” he writes.
• Your favorite creator's heartfelt political endorsement may be bought and paid for — and the FEC lets it stay secret. Max Weber, a New York City-based political and communications strategist, on the loophole laundering dark money and influence through parasocial relationships and what can be done to close it.
• Tyler Whitmer and Ben Rashkovich of Legal Advocates for Safe Science and Technology explain xAI's suit against California's AB 2013, a law requiring AI developers to publish general documentation of their training datasets—what they call the functional equivalent of an ingredient label.

More on AI governance, from local to global

• “The UN’s global AI governance discourse appears to simultaneously throttle the gas and brakes on AI, promoting global AI adoption while it warns about the serious risks it poses to society,” writes Bradley McNeil, a PhD Candidate at McMaster University. Its AI vision risks entrenching the very concentration it warns about, he writes.
• James Görgen, advisor at the Ministry of Development, Industry, Foreign Trade and Services (MDIC) of Brazil, finds three notable absences from the Stanford "We Must Act Now" letter: those responsible for AI's risks, the Global South, and the governance race already underway—from Washington's executive orders to China's new WAICO.
• OpenAI has discussed giving the U.S. government a 5% equity stake. India may take a stake in Sarvam AI. Gulf states are building AI through sovereign wealth capital. S. Yash Kalash, senior fellow at the Centre for International Governance Innovation (CIGI), argues the nationalization of AI has already quietly begun.
• AI reached local governments before policy did. Pitt Cyber’s Elise Silvaand Beth Schwanke and Local Government Academy’s Joy Ruff warn that if adoption keeps outpacing governance, the potential costs—compromised data, biased outputs, eroded trust—land on governments least equipped to absorb them. “Small jurisdictions like those represented in our survey make up the overwhelming majority of American governments, yet they get the least attention in AI policy and governance,” they write.
• The hype says Chinese open-source AI is catching up. Rollins College professor JJ Jasser's asserts Kimi K3 is not open-source. It's "open-washing"—borrowing openness's credibility while withholding the transparency that justifies it.

Building what’s next

• "Context is a new, developing layer of the internet," write Mallory Knodel, founder and director of the Social Web Foundation, and FABRIC co-creators Evan Friedman and Brad Friedman. They make the case for building it as a shared standard, not a feature any one company owns.
• Independent prosocial design research can shape what gets built, but researchers lack the pathways and incentives to produce it. Audrey Hingle, who leads communications work for the Prosocial Design Network, and Julia Kamin, managing director at Prosocial Design Network, on what needs to change.
• For the first time, a brain implant has been approved for routine clinical use outside a trial. Ebani Dhawan, a technology policy analyst and science writer, on what the approval process did not cover and why that matters.

What we’re watching

Apparently unhappy with the reaction to his interview with The Economist’s editor, Zanny Minton Beddoes, billionaire Elon Musk has now labeled her a “traitor to the West.” Tech Policy Press readers will not be surprised by any of the alarming, bigoted and combative things Musk said in the interview, but his reaction says something about his psychology at the moment.

On Friday, the European Commission announced its “preliminary findings indicating that TikTok accounts of minors do not meet the safety standards required under the Digital Services Act (DSA).”

If you have analysis or commentary on these or other issues, learn how to contribute.

I wish you the best for the week ahead!

-Justin