The US and China Also Discuss Limits for AI

The US and China opened high-level talks in New York on September 20, ahead of the summit between Donald Trump and Xi Jinping. US Treasury Secretary Scott Bessent and Chinese Vice Premier He Lifeng discussed the economy, trade, and critical minerals, as well as AI safety. According to Reuters, among the ideas discussed were possible "guardrails" for AI — mechanisms and limits meant to reduce certain risks posed by advanced systems. The talks also covered the risks of systems capable of improving themselves and of technologies with both civilian and military uses.

Why it matters: AI safety has thus landed directly on the agenda of one of the world's most important geopolitical relationships. For ordinary users, the discussions may seem distant, but the rules set by major powers can later influence what kinds of models can be developed, how they are controlled, and what restrictions will exist for certain uses. At the same time, the two countries are trying to preserve their technological edge in the competition.

Trump Announces an "AI Force" and an "AI Czar"

Donald Trump announced on September 19 that he will appoint a new artificial intelligence advisor, called an "AI czar," and that he intends to create a structure called "AI Force." According to Reuters, Trump has not yet provided details about the composition or responsibilities of this structure. The US president has backed rapid AI development and indicated he prefers using the existing legal framework for issues related to the technology's use, rather than new federal regulations.

Why it matters: The announcement shows that AI is being treated in Washington not just as an industry, but also as a strategic matter. It remains to be clarified, however, what "AI Force" will actually do, which institutions will be involved, and what powers the new "AI czar" will have. The initiative comes at a moment when the US administration is simultaneously discussing AI security, competition with China, and the level of regulation needed for advanced technologies.

"Your AI Assistant Wants Your Password" — Personal Agents Are Getting More and More Access

AI agents are trying to move beyond simple conversation into taking actions on the user's behalf: they can manage emails, make reservations, buy products, or interact with other services. The problem is that such tasks may require access to accounts, documents, passwords, and payment information. According to The News, security experts recommend that users start with the most limited permissions possible. Facebook's former chief security officer, Joe Sullivan, recommends that users "start narrow," giving an agent only the access needed for a given task and expanding it later if the system proves trustworthy.

Why it matters: An agent that can access email, use a card, or modify account information can produce direct consequences. In this case, convenience and security become part of the same equation: the more an agent can do on a user's behalf, the more opportunities it has to make a mistake or be manipulated. For the average user, granting only the minimum access necessary becomes one of the simplest protective measures.

An AI Ring Can Record Meetings, Classes, and Conversations — But It Raises a Privacy Problem

The company Vocci is proposing a new type of AI device: a ring that can record conversations and meetings, then generate transcripts and summaries. According to TechCrunch, the device costs $249, weighs less than six grams, and can record for up to about eight hours. Users can start recording with a tap, and can later use AI to search for information within the recorded conversation.

Why it matters: The ring turns a function once associated with phones or voice recorders into one that can be worn almost permanently. This is where the consent problem arises: the device looks like an ordinary piece of jewelry, and the light signal indicating that recording is underway can be hard for others to notice. For meetings, classes, or private conversations, the question becomes who knows that the conversation is being recorded, and where that recording ends up.