Latest / UpNext AI / Europe’s AI Sovereignty Push, Asia’s Export-Control Opening, and Faster AI Bug Hunting | UpNext AI – June 29, 2026
Transcript
- Welcome to the UpNext AI podcast. It's Monday, June 29th, 2026, and here's what matters in AI today. First up, Europe’s AI sovereignty push is getting louder. In a reported piece from WIRED, the core story is simple: Europe wants its own AI. But the same piece also says it’s still a stretch to think the continent can build a top-tier model on par with the very best labs. That tension is really the story here. Europe sees the strategic risk clearly, but turning that concern into a genuinely competitive model stack is a much harder job. According to the article, that urgency was all over the conversation at Vivatech in Paris, where leaders worried about getting stuck with American AI trained on American values. The piece says the word of the moment was sovereignty, and not in an abstract sense. It was about whether governments, companies, and institutions in Europe can rely on foreign model providers for core infrastructure. WIRED also points to a funding gap. One figure repeated in the piece was that Anthropic’s recent 65 billion dollar fund-raise was larger than the total invested in European and UK AI startups last year. The article says Europe is trying to answer that with new funding, collaborations, and potentially less resource-intensive next-generation approaches. And then there’s the political angle. The WIRED story argues Europe has an unusual advantage right now: Donald Trump. More specifically, it says the Trump administration’s policies are making dependence on U.S. AI feel less safe and less stable. The piece describes that as a wake-up call, especially after export-control moves around advanced U.S. models. In other words, Europe’s sovereignty push is not just about pride or industrial policy anymore. It’s increasingly about continuity of access. So the bottom line on this one: Europe clearly wants its own AI future, but by WIRED’s own framing, the ambition is ahead of the current capability. The strategic pressure is real. The question is whether it can translate that pressure into actual frontier capacity. And that leads directly into the second story, because the market is not waiting around. TechCrunch reports that new models are launching in Asia that promise Mythos-like capabilities without fear of an export ban. The article frames this as a direct response to the uncertainty created by U.S. restrictions around Anthropic’s most advanced models. One example in the story is Sakana AI in Tokyo, which launched a model called Fugu. According to TechCrunch, the company says the model stands shoulder-to-shoulder with leading systems like Anthropic’s Fable 5 and Mythos Preview, and is designed for agents with the ability to orchestrate access to other models through their APIs. TechCrunch also reports that Sakana is openly advertising frontier capability without the risk of export controls. The same piece says Chinese firm 360 reportedly unveiled Tulongfeng, an AI tool it says can go head-to-head with Mythos, along with another tool called Yitianzhen for cyber defense and incident response. What matters here is less the brand names and more the market signal. TechCrunch says U.S. AI labs may never recover this enormous market if customers decide they can’t risk building on tools that might disappear behind policy barriers. The article notes that Anthropic said its run-rate revenue crossed 47 billion dollars in May 2026, though it also says the company’s Asian enterprise exposure isn’t publicly known. Still, TechCrunch’s reporting suggests that at least two companies, one in Tokyo and one in Beijing, have already stepped into the gap left by the U.S. restrictions. And even if those restrictions eventually ease, the article argues local alternatives trained for local language and nuance are already gaining ground. So if the Europe story is about sovereignty as long-term strategy, this Asia story is sovereignty as immediate market opportunity. Now to the research pick. A paper published June 26th looks at using a large language model-driven framework to find business-logic vulnerabilities in power-system microservice APIs. The title is a mouthful, but the practical idea is straightforward. Business-logic vulnerabilities are flaws in how a system is supposed to behave, not just simple coding mistakes. That makes them harder for traditional security tools to catch, because those tools are usually better at spotting technical misconfigurations than understanding process rules and state changes. In this paper, the researchers propose a framework called LLM-SID. According to the abstract, it models business semantics, formalizes state transitions, and uses a multi-agent system to mine business logic constraints and generate proof-of-concept scripts. The paper reports an 88.0 percent recall rate and 84.6 percent precision on representative microservice benchmarks. And the headline number is the time reduction: it says the framework cut manual testing time from an average of 4.5 hours to 3.2 minutes per vulnerability. This is specific to power-system microservice APIs, so it’s not a blanket claim about all software security. But it is a useful sign of where LLM-based security tooling may be most valuable: not replacing every scanner, but helping teams reason through complex workflows that older tools tend to miss. Bottom line: if the result holds up in broader testing, this is the kind of targeted AI security use case that could matter quickly in critical infrastructure environments. ...Are you building apps with voice? Elevate your app's voice capabilities with ElevenLabs. Their API is a game changer for embedding dynamic, responsive voice interactions in your applications, providing unprecedented realism, flexibility and latency. In fact, you're listening to one of their voices - right - now. If you are a developer looking to elevate user experience with natural voice interfaces, this is your solution. Visit up next dot fm slash eleven to check out their latest offerings. ... OpenAI says it’s beginning a limited preview of the GPT-5.6 series. According to the announcement as quoted by Simon Willison, the lineup includes Sol as the flagship model, Terra as a balanced model for everyday work, and Luna as the fast, affordable option. The same post says Terra is positioned as competitive with GPT-5.5 while being cheaper, and that broader availability is planned in the coming weeks. Also worth noting, Interconnects’ latest Open Artifacts roundup argues the open ecosystem is expanding in breadth. This edition focuses on Zyphra, Cohere, and Poolside, and the main takeaway is that open model development is no longer being driven by just one kind of company or one geography. That’s more ecosystem signal than breaking news, but it’s a useful read on where momentum may be shifting. And finally, Simon Willison highlighted a useful security stress test: 2,000 people tried to hack an AI assistant by emailing an OpenClaw test instance in an attempt to leak secrets. The challenge reportedly generated 6,000 attempts and involved a 500 dollar reward, and no one managed to extract the secret. That doesn’t mean prompt injection is solved, but it does suggest some frontier-model defenses are getting harder to break in simple attack setups. Before we wrap up, a quick note: this podcast is generated with the assistance of AI and is intended for informational purposes only. All referenced articles, research, and commentary remain the property of their original authors and publishers. If you enjoyed this episode, don't forget to subscribe, rate, and leave us a review! And that's your briefing for today. Full source links are in the episode notes, and we'll be back tomorrow with what's up next!