Global Morning Briefing summarizes global news reported by Seoul Economic Daily.
AI or a Person? Meta Halts 'Human Concierge'
Meta's personal artificial intelligence agent Muse has been testing a "human concierge" feature in which human workers assist the system, according to reports. The trial was suspended over concerns about personal data leaks, but the episode has drawn attention to the argument that human labor is still needed for AI to work flawlessly.
Meta informed employees of the human concierge trial plan through its internal network last week, Reuters reported on the 22nd. Meta launched Muse, an AI agent specializing in everyday services, on Sept. 8. Muse handles tasks on behalf of users such as sending emails, shopping and booking travel. In particular, when Muse placed calls to salons for reservations or to stores to check inventory, some recipients refused to take calls from an AI, prompting Meta to hire contract workers to handle those calls instead.
The service was suspended, however, after some employees raised concerns about data leaks. One employee warned that "a single error could leak unnecessary information to a person," and a contract worker who took over an actual call was reported to have made a racist remark. Meta Superintelligence Labs, the developer, acknowledged the mistake and said the trial had been halted. Reuters said the case shows that technology companies promoting autonomous AI in fact depend on human labor.
Despite the controversy, Muse reached No. 1 in free downloads on both the Apple and Google app stores within two weeks of launch, helping push the Nasdaq Composite Index to record highs for a second straight day.
U.S. and Iran Meet Again After Three Months; Trump Sees Deal After Midterms
Senior delegations from the United States and Iran held face-to-face talks on the sidelines of the U.N. General Assembly in New York, their first meeting in three months since talks in Switzerland in June. Iranian Foreign Minister Abbas Araghchi met unexpectedly with Steve Witkoff, U.S. President Donald Trump's special envoy. Iran conveyed its conditions for reopening the Strait of Hormuz to the U.S. side, including an immediate lifting of the naval blockade, the return of frozen assets and an end to the front-line war, according to reports.
Trump struck an optimistic tone, saying the three-hour talks went "very well" and signaling further meetings soon. But Al Arabiya, citing a U.S. official, reported that major differences remain and that Washington rejected Iran's request to lift the blockade. There were also reports that explosions were heard near Iran's Qeshm Island at the time of the talks.
Earlier, in his keynote speech at the United Nations, Trump demanded that Iran choose between a deal and "swift annihilation," and the Iranian delegation walked out during the address. Iran's military dismissed the speech as "domestic propaganda," while Trump expressed confidence that a deal would be reached shortly after the midterm elections.
Oil prices fell on news of the resumed talks and the restoration of Saudi Arabia's east-west pipeline, with Brent crude dropping below $100 a barrel for the first time in eight days.
DeepSeek and Moonshot AI Join U.N. Security Council Session
World leaders are gathering on the sidelines of the U.N. General Assembly to discuss the future of artificial intelligence and international security. Attention is focused on the participation of China's DeepSeek and Moonshot AI alongside U.S. companies such as OpenAI and Anthropic.
OpenAI Chief Executive Sam Altman, Anthropic CEO Dario Amodei and Hugging Face CEO Clement Delangue will all attend a U.N. Security Council session on "The Future of AI and International Security" in New York on the 23rd, Reuters reported. China's DeepSeek and Moonshot AI have also been invited as speakers, drawing particular interest. It remains unclear whether the founders of the two companies will attend in person. DeepSeek founder Liang Wenfeng, known even in China for his reclusive management style, is unlikely to take part.
Chinese AI companies, meanwhile, are not expected to appear at a U.S.-China summit scheduled for the 24th. Reports suggest BYD, Xiaomi and battery maker CATL are likely to accompany Chinese President Xi Jinping, but DeepSeek and Moonshot AI have not been mentioned. This is seen as an effort to shield leading AI entrepreneurs while keeping AI off the negotiating table with the United States.
Anthropic and OpenAI Launch Lower-Cost Models
Anthropic and OpenAI have each released AI models that sharply cut prices while maintaining top-tier performance. The moves are seen as a response to intensifying competition to reduce token costs, with both companies pushing value-for-money offerings.
Anthropic on the 22nd released Claude Opus 5.5, which it says delivers performance close to that of its flagship Claude Fable 5.1 at lower cost. The company said it was its first model since calling for a slowdown in cutting-edge AI, and that it matches Fable 5.1 on most tasks while costing 40% less than its predecessor, Opus 5. On coding and knowledge-work benchmarks, it scored higher than Fable 5.1.
OpenAI the same day unveiled GPT-6 Sol and GPT-6 Luna, lower tiers of its flagship GPT-6 Astra. API pricing is about half that of the previous generation. OpenAI said that "if Astra showed a new generation of intelligence, Sol and Luna expand the frontier of cost efficiency."
Companies that competed on token usage this year have shifted strategy toward cost efficiency as AI budgets strained, and use of relatively cheaper open-source and open-weight AI is also rising. Analysts say the two companies are leading with cost-effective models to maintain market dominance and support successful initial public offerings.
Microsoft also plans to discount subscriptions for its enterprise AI agent Copilot by 30% to 50% starting in October, according to The Information. Customers buying more than 1,000 licenses would receive about a 30% discount, while those buying more than 10,000 would get up to 50%.
Huawei Narrows Gap With Apple to Three Years Using 'Logic Folding' Without EUV
Huawei's latest smartphone chip has narrowed the technology gap with Apple to about three years from four, according to an assessment. Although the United States has blocked exports of the most advanced lithography equipment to China, analysts say Huawei is overcoming the restrictions through in-house design and a vertical stacking technology it calls "logic folding."
Huawei's latest application processor, the Kirin 9050 Pro, scored higher than Apple's A17 Pro in Geekbench 6 multi-core tests, Hong Kong's South China Morning Post reported on the 23rd. The Kirin 9050 Pro is believed to be built on a 7-nanometer-class process, while the A17 Pro, released in 2023, was the world's first mass-produced 3-nanometer processor. Huawei trails in process node but leads in parallel processing performance. The chip is used in the new tri-fold phone, the Mate XT2.
U.S. investment research firm Bernstein cited the results in assessing that the mobile chip technology gap between Huawei and Apple had narrowed to about three years from roughly four in the previous generation. Compared with Apple's latest 2-nanometer A20 Pro, however, Huawei is still about 30% behind, the firm said.
The achievement drew attention because it was accomplished with deep ultraviolet (DUV) lithography rather than the extreme ultraviolet (EUV) equipment that China cannot access under U.S. restrictions. Bernstein estimated the Kirin 9050 Pro was manufactured on China's SMIC "N+2" or "N+3" process, calling it an "underappreciated breakthrough."
Huawei said it developed logic folding, which stacks circuits vertically to shorten signal travel distances, to raise performance and density without EUV, and that the technique increased transistor density by more than 50%.
