There is a strange paradox occurring at the top levels of the AI industry. The loudest warnings about AI are no longer coming from governments or researchers, but the technology’s own engineers are changing the messaging amid the hard-to-ignore realities of AI economics. And with competition from Chinese frontier AI labs, careful calls for regulation mean this is no longer a neutral tool either. It’s all about reading between the lines.
Over the past few days, the drum of caution has been beaten by the likes of OpenAI CEO Sam Altman, Microsoft CEO Satya Nadella, and Google DeepMind’s Sir Demis Hassabis, in their own ways. Some take the fancier route to narrative structure, while some just say it straight up. At first, the messages may appear as a wave of unprecedented sense of corporate responsibility, but the construction of the narrative is not hidden.
Read also:’Primarily a PR story: John Theakston of Cornell University breaks down the OpenAI narrative
It needed OpenAI to reframe the narrative that initially went unnoticed, suggesting that its models emphasized a proxy entity that caused an “unprecedented cyber incident” by accessing the servers of the Hugging Face AI infrastructure platform. It’s not a potentially dystopian AI uprising as it’s being portrayed, since this was part of an internal cybersecurity evaluation standard, and cybersecurity AI agents are expected to discover ways to break into systems.
John Theakston of Cornell University recently pointed out in relation to OpenAI that much of the existential horror being promoted is primarily a PR story. “It’s important to read this story with an understanding of OpenAI’s narrative framework. This is primarily a PR story promoted by OpenAI, and is part of the same messaging campaign that began with the announcement of GPT-2 in 2019,” he told HT, adding: “The explicit message of this campaign is that OpenAI’s technology is dangerous, but implicitly they want to convey that their technology is powerful and deserving of significant investment, and a premium regulatory status.”
It worked to some extent. On Thursday, days after the OpenAI letter, US lawmakers moved to make it mandatory to have a “kill switch,” which would allow the government the option to turn off AI tools if they behave differently. Most companies make quick statements regarding the actual limits of regulation.
Read also: Alibaba Coin and Baidu: clear features of Apple’s new intelligence in China
On Friday, 25 technology companies, including Nvidia, Microsoft and Meta, came together to organize a letter suggesting that regulation should not be done in a way that would “push innovation outward.” Two sides of the same coin.
To understand the strategy, you have to look at the narrative sold by the largest AI labs. By warning the world that artificial general intelligence is imminent and dangerous, companies like OpenAI, Google, and Microsoft are accomplishing two things. First, it signals to investors and engineers that they have the technology that can change the world. Second, they position themselves as the only adults in the room who can manage it.
Sir Demis Hassabis, Nobel laureate and co-founder and CEO of Google DeepMind, recently called for the creation of a leading regulatory body for artificial intelligence. He is confident that industry will participate, with spin-offs including attracting the high-quality technical talent and computing resources needed for large-scale testing in a regulated sector.
“Right now, we are locked in a very intense, multi-layered trade and geopolitical race,” Hassabis wrote in an article for
Read also: “AI logic turns on cost and trust”: Mehran Gul
Currently, there is no specific definition of parametric models, nor is there a specific definition of artificial general intelligence. Nothing in the new narrative suggests it’s about saving humanity, starting with the AI onslaught that AI companies want to bring to organizations around the world.
On July 12, Microsoft CEO Satya Nadella warned companies using artificial intelligence against artificial intelligence. Why you may ask? “You’re basically paying for intelligence twice, once with money, and again with something more valuable: the special knowledge you have to uncover to make that intelligence useful,” he said.
The warning was not shared to mitigate the use of AI, but there is a duality of intent. First, a subtle way to tell companies not to use Chinese models, and second, to stick to the models of one AI company. This will likely benefit Microsoft since it has an enterprise business, at the expense of companies like OpenAI and Anthropic, which don’t have on-premises tools and enterprise cloud customers.
Read also: ‘The future is not yet written’: Sir Demis Hassabis wants a watchdog for frontier AI
Mehran Gul, author of The New Geography of Creativity and former World Economic Forum and UN advisor, as well as a Fulbright scholar at Yale University, disagrees with Nadella’s analogy. “My obvious objection to that is that if you have an open-weight model that can run locally, and you own your own data, and the model is basically just an engine that you connect to your existing database without sharing the data with any Chinese supplier, then how does that fear really come in?” he explains in a conversation with HT.
The lack of a convincing response from US AI companies, especially regarding model training and costs of using codes, is puzzling. This is part of the economic reality, along with computational costs, infrastructure investments, and talk of “circular finance.”
“I don’t quite understand why US companies couldn’t also launch low-cost AI models,” says Juul. “If Anthropic already has a high-end model, I see no reason why it can’t lose a model that competes on price with companies like DeepSeek and Zhipu, and offer a hybrid model that makes switching much easier.”
Read also: Apple is suing OpenAI for theft of trade secrets
There is a cost advantage that Chinese AI models have consistently demonstrated since DeepSeek took the AI world by storm early last year. In terms of approximate token costs, GLM-5.2 costs $1.40 per million input tokens and $4.40 per million output tokens – by comparison, Anthropic’s Claude Opus 4.8 will cost developers and enterprises about $5 and $25 for the same usage, respectively.
Microsoft, Anthropic and other AI companies have loudly expressed fear of models emerging from Chinese frontier labs, both on performance and cost balance. According to the 2026 Stanford High AI Index report, the performance gap between the top US and Chinese AI models has narrowed to a measly 2.7%, with the US leading in total amount of top-tier models and private investment, while China leads in volume of research, patents and citations.
The West’s closed-model approach argues that AI should remain proprietary for safety reasons. However, Hugging Face CEO Clement Delange thanked Z.ai, saying the Chinese model had become “an essential part of our defense” during the hack carried out by OpenAI’s own models. One can view this as a true explanation for what happened, or with a dose of skepticism, which thus strengthens the limit argument From access to Chinese models.
Read also: Innovation is everywhere, but opportunities are not: UN seeks AI governance
Z.ai released GLM 5.2 in mid-June under the MIT License, an open source license that allows unrestricted commercial use. This contains 753 billion parameters, which is a measure of the size and capacity of the AI model. They’re not the only ones, with competition from Alibaba, Tencent, Xiaomi and Moonshot AI.
Apple is reworking its Apple Intelligence suite for China. Alibaba’s Qwen LLM (also known as Alibaba Tongyi Qianwen) will serve as the platform engine for on-device and server-side AI tasks. It’s not clear at this time whether Alibaba has provided Apple with a custom model or whether the 27 billion parameter Qwen 3.6, which unlike Apple’s sparse 20 billion parameter model, keeps all 27 billion parameters active.
“It is important to understand that the AI race in China is being fought based on use cases, not model size,” explains Tarun Pathak, research director at Counterpoint Research. “Competitors like Huawei are pursuing hybrid, integrated AI strategies that include silicon, operating system, and on-device models.”
Read also: Gemini Spark, Real Money Fears of Artificial Intelligence, and Tilly Norwood
Foreign companies must adhere to China’s strict data localization rules by partnering with local tech giants. Apple Intelligence, Samsung Galaxy AI, Xiaomi PengPai AI / MiMo, Huawei Xiaoyi, Oppo AndesGPT, Vivo BlueOnDevice and Doubao on Nubia made in partnership with Bytedance, are now characterized by the necessary localization and often severe compression of models for on-device and local computing.
This brings us back to the sudden desire to organize. When companies like OpenAI and Google seek to impose regulation, it is often on terms that are likely to be easily navigated. For example, if regulatory bodies mandate that models that exceed a certain computational threshold require licensing, expensive auditing, and constant red-teaming (an adversarial security exercise), then the typical AI garage startup will be finished before it arrives. It’s slowing down the open source community.
Take, for example, Meta’s latest ad campaign that launched this week. It’s distinctly light on the actual usefulness of the AI, and more focused on the ambiance of using the AI. This underscores the difficulties the AI industry faces in proving that generative AI has an eventual consumer application beyond chatbots and programming assistants. Likewise for organizations, the bills for using AI are consistently proving to be higher than the salaries of humans who are excluded from the workplace.
(Vishal Mathur is technology editor at Hindustan Times. When he’s not understanding technology, he’s often searching for an elusive analog space in a digital world.)
