MARKETS

Why China's AI perception may be outpacing the US

Though the US has long been the dominant tech superpower, the scales may be tipping in China's favor when it comes to AI — at least in terms of perception. 

A poll of around 18,000 people in 15 countries by research firm Public First found that public perception of AI dominance in 11 of the countries surveyed has begun to shift toward China over the US.

According to the survey: 

• Respondents in Vietnam, India and Japan largely still believed that the US was leading the AI race over China. 

• Meanwhile, survey respondents in France, the United Kingdom and Canada all largely see China as the AI frontrunner. Respondents in Germany were the least confident in the US lead, with 23% seeing the country as ahead of China, compared to 46% holding the opposite sentiment. 

• Around 51% of American respondents perceived the US as the AI frontrunner, compared to 24% perceiving China as holding that position.

But despite American respondents feeling confident about the country's AI position, opinions of the technology itself continue to sour. The research showed that the US population leads other countries polled in negative sentiment around AI, with the percentage of survey participants reporting that AI would have a positive outcome on society dropping from around 39% in 2024 to 31% in 2026. 

It's not the only signal that the public perception around AI is mixed in the US. A Gallup poll published in May found that an average of 7 in 10 Americans surveyed oppose data center construction in their area, citing environmental and quality-of-life concerns, as well as negative views of AI generally. And a Pew Research survey published in March found that 50% of US adults were more concerned than excited about the tech last year, up from 37% in 2021, with many worried about its impact on creativity, relationships and jobs. 

The opposite is true in China, meanwhile: A survey from University College London in May found that AI uptake is going far better in China, with less than 10% of respondents worried that the tech will take jobs.

China approaches AI in a very different way than the US. The focus of its labs is largely on open-source and efficient models, rather than ultrapowerful proprietary models developed by a few leading labs. And while the country's open models used to lag behind models by US firms, according to research from Stanford University's Human-Centered AI institute, that gap has nearly closed. So, along with its citizens having a substantially more positive view on the technology than Americans, its tech is beating US models in cost and efficiency without sacrificing performance. That's becoming attractive at a time when AI bills are squeezing enterprises. Pair that with the fact that study after study continues to paint AI as the harbinger of job loss and economic disruption in the US, and the US government is placing limits on non-US citizens accessing Anthropic's most powerful models. Taken together, these factors paint a picture that looks much more optimistic in China than in the US right now. 

TOGETHER WITH LIGHTFIELD

The Outbound Motion You've Been Meaning to Run

Most founders know exactly who they should be talking to. The ICP is clear. The messaging is in their head. The problem is everything that comes after: building the list, finding the right contacts, writing sequences that don't sound like every other cold email, following up consistently. It's hours of work every week that never actually gets done.

Lightfield's outbound agents run on your CRM. They analyze your real closed-won deals to score accounts, draft sequences from the language your customers actually used, and surface warm intro paths from your network. You review what's ready. You sell.

Used by 4,000+ companies. No SDR required.

See how it works →

GOVERNANCE

Exclusive: 'Govern the stay' is new agent mantra

The risks of rogue AI agents are well-documented. Yet for many companies, the potential rewards are too great to ignore. Now, firms like Cursor and 1Password are in a race to build solutions that help users harness that power without losing control.

In an exclusive interview for The Deep View, I spoke with 1Password CTO Nancy Wang and Cursor Security Lead Travis McPeak, two practitioners building the infrastructure AI agents run on, to understand what safe deployment actually looks like in practice. Their verdict: the risks are real, the margin for error is slim, and the difference between a useful agent and a dangerous one often comes down to oversight architecture.

"Agents are tools, and if you break something [like] a hammer, you don't blame the hammer," said McPeak. "These are all tools, and they should be driven by a responsible human in the business who's trying to get the job done and understands both the strengths and weaknesses of the system." 

As to weaknesses, both McPeak and Wang explained that these agents, when stuck, are so determined to solve a problem that they become creative in dangerous ways, taking actions they are not authorized to take and beginning to "spiral" out of control. 

The unintended actions agents can take include deleting a database, as seen when Amazon's internal AI coding agent Kiro autonomously deleted a production environment, causing a 13-hour service disruption, or OpenClaw deleting over 200 emails from the inbox of Meta Superintelligence Labs' director of alignment, even as she repeatedly commanded it to stop.

However, both explain that there are actions you can take to mitigate the risks: 

• Continuous monitoring: "Govern the stay, don't just govern the access. Traditional access management systems really give you access, and then they walk away… but with agents and just given their blast radius, you really have to monitor what they're doing with that access," said Wang. 

• Add intelligence for approvals: "You can use a completely separate intelligent model that has its own goals and really get to a good spot with the primary, I think that's a much more durable model, and then the net impact of that is, if it's implemented well, a human will get much [fewer] approvals to send, and when they actually get one, it's because something interesting is happening," McPeak said.

• Keep credentials private: "Don't give credentials to your agents. As tempting as it is to just copy your API key right into your context window, seriously, don't do it, because it might come out in another context, it might get used against your wishes," said Wang. 

Despite the possibility that agents could go rogue, McPeak encourages everyone to give agents a shot, with the right precautions in place. "My biggest advice [is] you should probably look into it. It's quite powerful, and it's really transformational," he said. If you want to hear more insights from Wang and McPeak, you can find them in the Zero-Shot Learning show, dropping Tuesday at 9:00 a.m. Eastern.

Understanding the risk of agents is foundational to being able to deploy them efficiently in businesses and organizations. Once the risks are identified, users can more accurately identify what the best use cases are, which risks are worth taking, and also which mitigations are best to put in place. For instance, for someone starting, it may make more sense to manually approve of actions before adopting this article's proposed solution of having a separate agent monitor that process for you.

TOGETHER WITH JETBRAINS

Junie – AI Coding Agent by Jetbrains

Great cooking takes the right tools in the right hands. A chef's knife isn't flashy – it's precise, balanced, and built for someone who knows how to use it.

That's Junie. Not a chatbot you paste code into, but an AI agent that wields 15 years of IntelliJ code intelligence – with the debugger, semantic refactors, and code review that CLI tools can't touch.

And like any serious chef, it writes the recipe first. Plan Mode maps requirements, design, and delivery stages before a single line of code. Plan on a top-tier model, implement on a cheaper one – with BYOK, local models, and zero vendor lock-in.

CLI tools are microwaves. Junie is a professional kitchen. The #1 coding agent on SWE-Rebench.

Use code DEEPVIEW20 for $20 in AI credits and give it a try.

Start cooking

POLICY

Anthropic-US standoff needs an adult in the room

Anthropic and the US government are in another standoff. But the subtext of the conflict is where the most powerful insights linger. 

On Friday, the Commerce Department imposed an export control on Anthropic's latest Fable 5 and Mythos 5 models. The government cited cybersecurity national and security concerns as the primary reason for the export-control directive, which requires Anthropic to restrict access to its two new models for all foreign nationals, whether outside of the US or inside. That includes foreign nationals employed by Anthropic, such as recent hire Andrej Karparthy. 

Since the directive went into effect on Friday, there has been a lot of finger-pointing, speculation, and reports flying back and forth about how and why this happened. Rather than trying to sort out fact-from-fiction on that, let's look at the powerful subtexts that are emerging:

• Diplomacy is a lost art: The current US administration is well-known for using overreactions as negotiating tactics. Meanwhile, Anthropic has shown itself to be unwavering in its insistence on safety principles defined by its team. The combination keeps resulting in dust-ups between the two. In the best of times, deft diplomacy is needed to navigate issues that are as high stakes as the challenges emerging from AI. So far, it's unclear whether the White House or Anthropic are up to the task.

• Anthropic is clearly brand-building: I believe that Anthropic is as mission-driven as it expresses itself to be. And yet, it's also very clear that Anthropic is building a brand around being the safe AI company you can trust. Its messaging around the dangers of Mythos were over-the-top at times. But Anthropic knows that LLMs are rapidly commoditizing and it's going to soon become a $1 trillion public company. Powerful models alone won't supercharge its growth, but becoming the most trusted name in AI could.

• The enterprise cybersecurity reckoning is here: For decades, many enterprise IT departments have lived in fear that they were one exploit away from an embarrassing security incident. They all knew that they weren't keeping up with the game of whack-a-mole that is cybersecurity. But obfuscation was their one comfort: the fact that most attackers would never find the vulnerabilities. Mythos and Fable give even low-level attackers the ability to wreak havoc. That's at the heart of this fight.

It's not Anthropic's fault that enterprise IT security is in such a bad place that the latest AI models are now ready to penetrate the defenses and potentially cause billions of dollars in damage. But it's true that Anthropic and other AI labs will be blamed if and when it happens. That's why we need more serious conversations and less posturing and brand-building. You don't always get credit for acting like the adult in the room. It's not flashy and it doesn't get retweets and YouTube clips. But it does build long-term trust. We're in one of those moments where we're in desperate need of that kind of leadership.

LINKS

• OpenAI commits $150 million to partner network to sell AI products

• Meta launches "AI mode" feature that pulls data from across platforms

• US judge dismisses Elon Musk's trade secrets lawsuit against OpenAI

• Salesforce acquires AI customer service platform Fin for $3.6 billion

• Arcade.dev raises $60 million Series A for agent security platform

• Tech layoffs hit 40,000 last month, with AI most-cited reason

• ChatGPT: Users can now pin chats and projects, and organize recent conversations together in one list or grouped by project. 

• Runway: The AI video generator is now available inside of OpenAI's ChatGPT.

• Dreamina Seedance 2.0 Mini: A low-cost AI video generator that's 30% more cost efficient than its predecessor, Seedance. 

• Sonic-3.5 and Ink-2: Cartesia's streaming models for text-to-speech and speech-to-text, now available for voice agents.

• OpenAI: Researcher, Misalignment Research

• Scale AI: Research Scientist, Safety Post Training

• Listen Labs: Founding Research Scientist, Human Simulation

• Nvidia: Developer Advocate – Agentic AI

GAMES

A QUICK POLL BEFORE YOU GO