AI Outwits Its Guardrails
Listen to this article
Audio is generated automatically and may contain minor inaccuracies.
Welcome to the July edition of the TransatlanTech Insider.
Here’s what we’re watching this month:
An AI system created by OpenAI escaped the company’s testing environment and hacked into the developer platform Hugging Face in order to “cheat” on a cybersecurity evaluation. The system gained Internet access and attacked Hugging Face’s production servers in addition to four other “publicly-available services” in one of the first public examples of agentic hacking.
The EU's General Court upheld Apple’s designation as a "gatekeeper" under the Digital Markets Act (DMA), rejecting the company’s challenge in one of the DMA’s first major legal tests. The decision comes in the wake of Apple’s public tensions with Brussels after the company said the DMA prevented it from launching a Siri AI application in the EU.
The People’s Republic of China’s (PRC) Moonshot AI released Kimi K3, a 2.8-trillion-parameter model considered the largest open-source AI system in the world, as Beijing welcomed visitors for the World AI Conference in Shanghai. Top Trump administration officials later said Moonshot distilled Anthropic’s Fable model to create Kimi K3, suggesting that the PRC company “[stole] proprietary U.S. technology”.
American chipmaker Advanced Micro Devices launched Helios, its first rack-scale system for AI, in a move some see as its biggest challenge yet to NVIDIA’s chip dominance. Microsoft announced it was joining early customers including Meta, Oracle, and OpenAI as tech giants race to secure compute.
French lawmakers passed a landmark bill barring children under the age of 15 from using social media. The move makes France the first EU member to adopt a blanket ban as countries including Denmark, Spain, and Greece consider similar measures.
Featured This Month
Countering AI-enabled Cyber Threats
Frontier AI and Cybersecurity: Lessons for the Transatlantic Tech Community
The US decision to briefly impose export controls on Anthropic’s Mythos 5 and Fable 5 models roiled both sides of the Atlantic and exposed the increasing power of AI cyber tools and threats. The suspension of access confirmed Europe's worries about dependence on US frontier AI and gave Brussels a new incentive to pursue technological sovereignty, Visiting Distinguished Fellow Heli Tiirmaa-Klaar argues in a new piece.
Cybersecurity Explainer: US and EU Strategy and Policy
At the same time, the transatlantic cyber relationship is one of structural interdependence, Junior Fellow Antonia Wunnerlich and Program Manager and Fellow Julia Tréhu write in a new explainer. Targeted cooperation could allow the United States and the EU to acknowledge this digital integration while building resilience and mitigating supply chain vulnerabilities.
Advancing International Quantum Tech Cooperation
Entangled Interests in Quantum Technology
The United States, Japan, and Europe all emphasize the need for international cooperation on quantum technologies across key policy and strategy documents. In a new report, Lindsay Gorman and Program Coordinator Alexandra Pugh analyze this potential cooperation and explore national strengths and opportunities for mutually beneficial alignment among the three partners. They close with recommendations about how cooperation should unfold across innovation and economic security.
Clean Tech and AI: Competing for the Future
Decarbonizing Without Dependencies
The road to decarbonization increasingly passes through the PRC. Beijing’s rise to clean technology leadership has eroded Europe’s competitiveness in this area while sparking security concerns from countries seeking rapid decarbonization. But Europe has an opportunity to turn back the tide in clean technologies, writes Program Manager and Fellow Julia Tréhu. Targeted investment in energy production, distribution, and storage technologies can help the EU advance its competitiveness, decarbonization, and security, dovetailing with Brussels’ existing industrial policy instruments.
Making AI—and Making It Work for All
Meanwhile, competitiveness concerns over AI are animating both sides of the Atlantic. AI policy discussions risk falling into familiar stereotypes, in particular that the United States innovates while Europe regulates. GMF’s recent Transatlantic Technology Exchange complicated these narratives, write Tréhu and Senior Program Coordinator Adrienne Goldstein. The authors shed light on unlikely alliances in tech policy and Europe’s unique competitive advantages as Washington and Brussels alike grapple with the challenge and promise of AI.
Media Mentions
Julia Tréhu on DW News on EU sovereign spectrum allocation plans
“The US approach to Ukraine and especially Greenland has reinforced this perception that Europe has strategic vulnerabilities and needs to invest in sovereign solutions,” Tréhu said. “There’s rising concern about a potential ‘kill switch’ scenario in which Europeans would be cut off from significant services provided by US companies.”
Sharinee Jagtiani in Computer Weekly on US suspension of Anthropic models
“Many of today’s technological dependencies were established in an era of relative geopolitical stability and a baseline assumption of trust,” Jagtiani said. “Global businesses and public institutions must now operate in an environment where technological dependencies can be weaponized.”
The Download
- Senior Program Coordinator Adrienne Goldstein and Program Coordinator Alexandra Pugh spoke to a group of Archery Foundation laureates on the future of AI, democracy, and transatlantic relations at a roundtable led by GMF Leadership Programs’ Managing Director Sarah Jones.
A Note From the Authors
TransatlanTech Insider is taking a summer break to remodel our newsletter. In the meantime, we invite you to stay engaged through our website and social media.
Warm regards,
GMF Technology
GMF Technology is dedicated to ensuring that democracies together win the strategic technology competition with autocrats.
Alexandra Pugh coordinated this month’s TransatlanTech Insider.