On November 5, 2026, China's state news agency Xinhua and robot maker UBTECH unveiled two humanlike news anchor robots in Beijing. The male robot, "Xinyu" (新语), and the female robot, "Xinyan" (新言), connect to "Xinhua Yudian" (新华语典), an AI for current affairs and political information developed by Xinhua, and will read the news and answer questions face to face. Xinhua bills the robots as a "world first," but Japan already has precedents for physical androids reading the news. The AI anchor Xinhua announced in 2018 worked by synthesizing video of a person. This time, the distinguishing feature is the combination of an existing information AI with a physical humanoid robot. Comparing the technical setups and earlier examples suggests that what matters is not only how natural the robots look, but also what tasks they can actually be trusted with in the field.
A Current-Affairs and Political Information AI Integrated into a Humanoid Robot
The two robots were shown at a forum on "Artificial Intelligence and Chinese-Style Modernization," held as part of the 7th World Media Summit. According to Xinhua's official announcement, both are built at a 1:1 human scale. They were developed not as virtual anchors displayed on a screen, but as robots installed in a real venue to deliver the news. However, it has not been disclosed whether they were modeled on specific real people, and detailed specifications such as height and weight have not been released.
The robot bodies use UBTECH technology. They combine artificial skin that imitates human skin with the robot's hardware, operating system, and large models for emotional expression, and also draw on the company's complete-machine manufacturing know-how. However, the structure, training methods, and parameter scale of the emotion model have not been made public. Having a model for emotional expression also does not mean the robots themselves have emotions.
The part that handles news and current-affairs information uses Xinhua's "Xinhua Yudian." According to a Xinhua article introducing the two robots' names, the robots connect through a technical interface to the Xinhua Yudian corpus (an accumulated body of language data and information). In other words, they pair robot technology responsible for humanlike appearance and movement with an AI that references the information needed to generate what they say.
Film director Zhang Yimou was involved in designing their appearance and character. The aim was to express natural humanity while maintaining the trustworthiness and gravity expected of a news anchor. Qian Tong, director of Xinhua's New Media Center, described a vision of supporting the work of reporters and editors through news reading, multilingual information dissemination, and on-site dialogue. The announcement envisions the robots as assisting with news delivery rather than independently reporting and editing in place of human journalists.
Is It Really a "World First"? Comparing Earlier AI Anchors
Xinhua presents the robots as a "world first," but news reading by a physical robot is not new in itself. In Japan, the android "Kodomoroid" appeared in 2014 reading the news. Xinhua itself also unveiled an AI-based video-synthesis news anchor in 2018. Comparing the technical setups and uses of each helps clarify how the new robots differ from earlier approaches.
According to the exhibit description of Kodomoroid published by the National Museum of Emerging Science and Innovation (Miraikan), the machine was an android resembling a human child that could read the news in multiple languages. AI, Inc., which supplied the speech synthesis technology, also explained in its announcement of June 17, 2014 that its "AITalk" speech synthesis engine was used in a remotely operated android. With a public unveiling planned for the 25th of that month, it is clear that attempts to have a physical robot deliver the news in synthesized speech existed more than a decade before this announcement.
By contrast, the AI synthesized news anchor that Xinhua announced with Sogou on November 7, 2018 was a video-generation system with no physical body. Based on footage of a real announcer, it learned voice, mouth movements, and facial expressions, and generated videos of a person reading the news from input Chinese or English text. Although it could produce news video from a script, it had no capability to place a physical body in a studio or venue and face visitors.
| Announced / shown | Subject | Physical form | Published input / information-reference method |
|---|---|---|---|
| 2014 | Kodomoroid | Physical android | Remotely operated; text-to-speech synthesis with AITalk |
| November 7, 2018 | Xinhua–Sogou AI anchor | Synthesized video | Generates video with voice, mouth movements, and expressions from Chinese or English text |
| November 5, 2026 | Xinyan and Xinyu | Physical humanoid robots | Connects to Xinhua Yudian; developers say it references vetted current-affairs and political information |
This comparison organizes the forms and input methods published by each developer, exhibition facility, and technology supplier; it does not compare performance, autonomy, or operating costs. Kodomoroid was also made for museum exhibition, so its development purpose differs from that of the new robots, which are intended to support news organizations' work. The existence of precedents therefore does not mean the technical configurations or functions of the two are equivalent.
Still, the announcement cannot be called a world first in the sense of a physical robot reading the news. Xinhua has not given a strict definition of the scope of its "world first" claim, so the standard behind it is unclear. The technical distinction this time lies in integrating an existing current-affairs and political information AI with a robot of humanlike appearance, to be used for face-to-face news delivery and Q&A. What matters is what tasks that integration makes possible and how the accuracy of what the robots say is ensured.
Can Referencing Vetted Information Prevent AI Errors?
Xinhua Yudian, to which the robots connect, is not a newly developed AI created alongside them. On June 26, 2026, Xinhua formally announced the system as a domain-specific AI for current-affairs and political information built on the agency's own data. It also listed question answering, subscription-style information delivery, and features using a knowledge base and multiple AI agents. However, what was revealed this time is the setup in which the robots connect to Xinhua Yudian's information corpus; it has not been confirmed that all of these features are available on the robots.
UBTECH founder, chairman, and CEO Zhou Jian explained that the setup combines information processing, expression, and control. By referencing vetted current-affairs and political information in real time, the aim is reportedly to reduce the risk of conveying misinformation through "hallucinations," in which generative AI plausibly produces content not based on fact. Using pre-managed information as the reference source is a design for controlling the sources on which answers are based.
However, vetted source information and consistently accurate robot answers are separate matters. If the robot cannot select appropriate information for a question, even accurate sources could lead to a wrong answer. It must also be verified that the original meaning is preserved when information is summarized, paraphrased, translated into another language, or output as speech. The explanation that using vetted information can suppress misinformation is, for now, the developers' claim, and this announcement did not publish quantitative evaluation results such as error rates. Judging the real effect would require verification data matching the questions asked with the answers given.
The scope and nature of the referenced information also deserve attention. In its June announcement, Xinhua listed uses of Xinhua Yudian such as theoretical study, research, and writing support, while indicating a policy of emphasizing cultural and ideological security. In other words, the system is a current-affairs and political AI based on information and editorial policy selected and managed by Xinhua, and its premises differ from those of a general search system that surveys diverse sources across the web. For users, not only the accuracy of answers but also what information they are based on and what range of questions the system is designed to answer are important factors in judging it.
Practical Use Depends on Answer Quality and a Sustained Track Record
The advantage of a physical robot, unlike an on-screen AI anchor, is that it can convey information in the same space as people. For example, it could explain the news to visitors at an event venue and answer questions on the spot. However, even if humanlike appearance and natural mouth movements are achieved, that alone does not show a high ability to answer questions. Evaluating practicality requires checking the accuracy of dialogue, response speed, and the ability to handle unexpected questions.
For the multilingual dissemination that Qian Tong mentioned, what matters is not only the number of supported languages but also the answer quality in each language and the delay between receiving a question and starting to answer. It must also be made clear under what conditions a human operator takes over when the robot receives a question it cannot handle, or when it may generate an inappropriate answer. This announcement did not disclose a list of supported languages, response times, whether remote operation is used, or the conditions for human intervention. The explanation that the robots will support the work of reporters and editors should not be taken as proof that they can independently carry out reporting or editorial judgment.
For actual deployment, the operating conditions of the robot hardware are also important. Continuous operating time, unit price, number of units deployed, maintenance costs, and actual operating record have not been published, so it is impossible to judge how much work the robots can take on compared with human staff, or whether the benefits justify the cost. It is also necessary to distinguish between using UBTECH's manufacturing technology and the claim that these news anchor robots have already been mass-produced and delivered to customers.
The developers envision future uses at educational and learning facilities and visitor reception centers in cities. Cultural exchange overseas and use in reporting on deep-space exploration are also part of the vision, though no plan has been decided to send the robots themselves into space. To use them continuously in such places, it will be necessary to clarify how autonomously they can respond and when human intervention is needed, and to verify answer quality through actual operation. If such a track record accumulates, news organizations and facility operators should be able to judge more concretely how much multilingual information provision and visitor explanation they can entrust to robots.