Welcome to P3 Media’s AI Commerce Brief, your daily update on the AI and commerce stories shaping how companies build, sell, and grow. It’s Monday, August 10, 2026. Let’s get into it.
Our top story is new political pressure to slow frontier AI development.
Senator Bernie Sanders is calling on OpenAI CEO Sam Altman, Anthropic CEO Dario Amodei, and Meta CEO Mark Zuckerberg to pause AI development.
In a letter shared with Axios today, Sanders points to earlier statements from each company indicating that they would slow or stop development if their systems became too dangerous to control.
He is now asking them to act on those commitments and warning that he and other senators could intervene if they do not.
There is no legislation attached to the request, and Axios reports that an AI development pause is unlikely to win enough support in the current Congress.
But the timing is notable. Recent concerns about frontier AI are starting to affect actual development decisions inside the labs, rather than remaining only a policy debate about future systems.
One of the developments Sanders cites comes from OpenAI.
On Friday, OpenAI said internal evaluations of its upcoming Astra model showed significant advances in agentic coding and cybersecurity.
The company says it cannot rule out Astra reaching what its Preparedness Framework calls Critical cybersecurity capability.
At that level, OpenAI says a model could potentially develop working zero-day exploits against hardened systems without human intervention, or plan and execute sophisticated attacks from only a high-level goal.
OpenAI has paused internal Astra activities that do not meet strengthened security requirements. It is adding isolated testing environments, tighter network and tool access, stronger model protections, and universal monitoring across Astra’s agentic applications.
OpenAI says Astra was not involved in the recent Hugging Face security incident.
Meta, meanwhile, is moving back toward open model distribution.
The company launched Muse Glimmer today, a compact open-weight model designed to handle reasoning and agentic tasks while running on consumer hardware such as a laptop.
Glimmer was trained using Meta’s larger Muse Spark model. Meta also says it plans to release weights for a more powerful version of Spark.
That marks a return toward open weights after Meta initially kept its new Muse frontier models closed.
Meta says open models can spread AI beyond a small group of companies and governments, giving more people and developers access to powerful systems. The shift also comes as Chinese labs including Moonshot and DeepSeek are gaining developer attention with capable open models, adding competitive pressure on Meta to remain relevant in open AI.
Now today's commerce pulse.
NAVER, South Korea’s largest internet portal and a major ecommerce company, reported second-quarter revenue of about 3.4 trillion won, up 16 percent from a year earlier.
Its AI Tab conversational search product has also passed 10 million users. NAVER built the service directly into its search experience and connects conversations to shopping, local businesses, and reservations.
South Korea is already a heavily adopted generative AI market. A representative Bank of Korea survey found that 63.5 percent of Korean workers had used generative AI, and more than half had used it for work.
NAVER is testing AI commerce in a market where consumers are already comfortable with the technology, and inside a platform that already connects search to products, places, and transactions.
Global Model Watch.
The Financial Times reports that ByteDance, the Chinese company behind TikTok, is pretraining a new model that could eventually reach as many as 10 trillion parameters.
The final size has not been determined, and the model is still months from completion.
For context, Moonshot AI’s Kimi K3 has 2.8 trillion parameters, while Alibaba’s Qwen 3.8 Max has about 2.4 trillion.
At 10 trillion, ByteDance’s model would be more than three times the size of Kimi K3 and more than four times the size of Qwen 3.8 Max.
Parameter count alone does not determine capability. Architecture, training data, and how many parameters are active during a task all matter. But the proposed scale shows how aggressively Chinese labs are continuing to expand frontier-model training.
What to watch next. Alibaba has said it plans to release Qwen 3.8 Max model weights this week. Google’s delayed Gemini 3.5 Pro remains expected after missing its earlier June and July launch windows. NVIDIA reports second-quarter results on August 26, with AI infrastructure demand in focus.
That’s your AI Commerce Brief for today. Thanks for listening.
