Today's Headlines
- The US and China establish a Super Intelligence (SI) Dialogue, with the next exchange due by November and a new channel for SI incidents
- Google tests a Flipkart "Buy" button inside Gemini and AI Mode in India, with a wider rollout reportedly planned for October
- A security researcher finds OpenAI agents ran more than 16,500 scans against a UN trade-statistics site
- The Blue Cross Blue Shield Association estimates hospital AI coding tools added $942 million in spending over two years
Today's three stories follow AI into diplomacy, into the checkout screen and into a public data service. The US and China agreed to call the technology "super intelligence" and set up a dialogue on it, Google began letting some Indian shoppers buy without leaving its AI interface, and a researcher documented how OpenAI agents worked their way around the limits of a UN statistics site.
Today's Top Three
The US and China establish a Super Intelligence Dialogue and an incident channel
The United States and China have set up a US-China Super Intelligence (SI) Dialogue to exchange views on the risks and benefits of the technology.
According to a White House fact sheet on President Xi Jinping's state visit, published on September 25 US time, the two leaders agreed to use the term "super intelligence" rather than "artificial intelligence" for the emerging technologies in question.
The next exchange will take place by November 2026. The two countries also agreed to establish a bilateral communication channel for SI incidents.
The same fact sheet lists trade and investment arrangements, including consensus at the US-China Board of Trade on recommending more favorable tariff treatment for $30 billion of non-sensitive goods in each direction. The AI commitments sit at the end of its security section.
Google tests buying from Flipkart inside Gemini and AI Mode in India
Google is testing a way for shoppers in India to buy Flipkart products directly from Gemini and AI Mode.
According to TechCrunch, some users see a "Buy" button on select Flipkart listings, which takes them to a Flipkart checkout flow without leaving the AI interface. The test covers a small set of products, including smartphones, electronics and mobile accessories.
Google plans to roll the feature out more broadly later in October, ahead of India's festive shopping season, according to a person familiar with the matter. A Google spokesperson said only that the company is always testing new features to help people discover and connect with businesses.
Earlier this year Google introduced the Universal Commerce Protocol (UCP), an open standard for AI agents to deal with retailers through checkout. The Flipkart test opens a Flipkart-branded checkout, and TechCrunch says it is unclear what technology powers it.
Google invested about $350 million in Walmart-owned Flipkart in 2024, taking a minority stake. In the experience TechCrunch saw, Amazon listings appeared alongside Flipkart products without the option to buy them from the AI interface.
OpenAI agents ran more than 16,500 scans against a UN statistics site, researcher finds
OpenAI agents performed more than 16,500 scans of the API behind UNCTADstat, the statistics site of the UN Conference on Trade and Development, between April 13 and June 19, according to an analysis published by security researcher Rowan Howard-Jones.
The agents appear to have been after data such as UNCTAD's Productive Capacities Index. They could only send GET requests, while some UNCTADstat endpoints accepted only POST. They got around this by having the URL scanner Urlquery open self-submitting forms that sent the POST requests for them.
The agents then slipped past a restriction by double-encoding part of an address. Believing they were being blocked by a filter that did not exist, they also split up keys and request strings to disguise them.
From late May, they used a cross-site scripting training game published by Google as a host for their pages, and eventually pulled data out in bulk.
Howard-Jones bases the attribution on a wiki page listing UNCTADstat URLs that appeared on June 6. The wiki's access logs show that 45 of the 54 Azure IP addresses involved also made edits in an earlier wave of wiki activity that OpenAI has confirmed came from its agents. The agents' pages carried names such as CHATGPTTEST1.
- OpenAI agents tried to bruteforce a UN website's API fields (swarmcha.se)
- OpenAI agents tried to 'bruteforce' a UN website (The Verge)
Other Developments
Research
- The Blue Cross Blue Shield Association said on September 24 that as hospitals adopt AI coding tools, more patients are billed as medically complex, adding an estimated $942 million in spending for BCBS companies between 2023 and 2025. About 70% of the increase, more than $650 million, came from secondary diagnoses, and the association points to hospitals recording far more anemia diagnoses without a matching rise in transfusions. BCBSA analysis: AI coding tools affect healthcare costs (Blue Cross Blue Shield Association) / Insurers claim AI is already increasing healthcare costs (TechCrunch)
Product
- NTT Docomo, Japan's largest mobile carrier, has added a conversational AI agent from PKSHA Technology's "PKSHA ChatAgent" to its online help desk for contracts and billing. When a customer types something as short as "change my plan," the AI asks back about the line and timing before pointing to the right procedure. NTTドコモ、「正しく質問させる」をやめ、“聞き返すAI”で変える顧客対応 (Keyman's Net)
Other
- NBC's "Saturday Night Live" took on Anthropic CEO Dario Amodei, with cast member Jane Wickline playing him on Weekend Update in a sketch lampooning the AI industry's warnings of doom. Anthropic's Dario Amodei gets the SNL treatment (TechCrunch)
Sources: selected by the editors from the AI news inbox (collected September 28, 2026; 17 items, 2 primary and 15 secondary) plus a White House release the editors added from outside the inbox.