OpenAI Halts Sora Development to Prioritise Enterprise Offerings

OpenAI is discontinuing its video generation tool Sora, the company said on Tuesday, in a surprise move by the ChatGPT maker, as it sharpens its focus on enterprise offerings ahead of a potential market debut later this year. The move involves winding down several Sora-related products.

Credit: UNSPLASH

"We're saying goodbye to Sora... we know this news is disappointing," the Sora team communicated in a post on social media platform X. Timelines for the app and application programming interface, along with details on preserving user work, will be shared at a later point.

OpenAI had published a blog post about Sora safety standards just a day before this announcement. Concurrently with Sora's discontinuation, OpenAI will also conclude its partnership with Disney, which was announced in Dec.

A spokesperson for Disney affirmed that the media giant respects "OpenAI's decision to exit the video generation business and to shift its priorities elsewhere." The three-year agreement had Disney set to invest USD 1 billion in OpenAI.

The partnership also permitted the artificial intelligence startup to utilise characters from Star Wars, Pixar, and Marvel franchises in Sora. A team at Disney was working with OpenAI's Sora team last night when the entertainment giant learned its partner was "pivoting strategy," according to a source familiar with the matter.

OpenAI first unveiled Sora in early 2024, astounding the world with software capable of generating feature film-like quality videos based on text prompts. This launch spurred AI companies across the U.S. and China to accelerate the release of their own AI video generation models.

The company launched the standalone Sora app in Sept. 2025, allowing users to create and share AI videos. These videos could be spun from copyrighted content and shared to social media-like streams. The Wall Street Journal first reported Sora's discontinuation earlier on Tuesday.

OpenAI CEO Sam Altman informed staff that the company would discontinue products that rely on its video models. Beyond the consumer application, OpenAI is also ending a version of Sora for developers.

The report further noted that video functionality would not be supported within ChatGPT either. This strategic shift arises as OpenAI faces intensifying pressure to ramp up its enterprise and coding product offerings.

Competition from rival AI startups and established technology giants is intensifying. Anthropic's concentration on training its models for coding has enabled its Claude Code product to achieve significant traction among developers, providing an advantage over competitors like OpenAI in the enterprise AI market.

---

• OpenAI is discontinuing its Sora video generation tool.

• The company is shifting its strategic focus to enterprise offerings and products.

• The discontinuation encompasses the standalone Sora app, a developer version, and video functionality within ChatGPT.

Source: REUTERS

NUS Researchers Develop Fastest Biohybrid Swimming Robot

Researchers at the National University of Singapore (NUS) have developed a new method for strengthening lab-grown muscles, leading to the creation of the fastest swimming biohybrid robot reported to date. This innovation addresses a significant limitation in biohybrid robotics.

Credit: NUS

The team built a platform connecting two engineered muscle tissues, allowing them to continuously pull against each other. As the muscles naturally contract during early development, they effectively create their own workout, removing the need for external stimulation or control systems.

This self-training approach resulted in record-breaking performance. The strengthened muscles powered a biohybrid swimming robot called OstraBot, which achieved speeds of 467 millimetres per minute.

The concept builds on the known biological behaviour of immature muscle cells spontaneously contracting as they develop. Researchers designed a system where two muscle tissues are linked through a sliding structure.

When one tissue contracts, it stretches the other, which then contracts in response. This creates a continuous cycle of motion that strengthens both tissues over time.

“As the cells mature, they naturally begin to contract spontaneously,” said Assistant Professor Tan Yu Jun. “Because the two tissues are connected, they continuously pull against each other, effectively exercising without any external control.”

The trained muscles achieved a maximum force of 7.05 millinewtons and a stress of 8.51 millinewtons per square millimetre, significantly exceeding typical lab-grown muscle performance. The method also utilises a widely available commercial cell line, which makes it easier to reproduce and scale.

The stronger muscles were integrated into OstraBot, a fish-inspired robot that mimics the swimming style of boxfish. Using a single trained muscle to drive flexible tails, the robot swam more than three times faster than versions powered by conventional muscle tissue.

Beyond speed, the system also demonstrated improved control. Researchers could adjust the robot’s movement by changing electrical signals and even trigger it to start and stop using sound cues such as clapping.

“The clap shows that the robot is not just alive — it is controllable,” Assistant Professor Tan stated. “In the past, muscle-powered robots either moved constantly without clear control or were too weak to respond visibly.”

Tan added that the strengthened skeletal muscle allows the robot to react clearly to an external signal, similar to how nerves control muscles in the body. The team believes this work removes a key bottleneck in the field and opens the door to more capable biohybrid systems.

The breakthrough could help unlock a new class of soft, efficient machines powered by living cells, with potential uses in medicine, environmental monitoring, and biodegradable robotics. Researchers are currently working on fully biodegradable robots that can safely break down after completing their tasks, including temporary medical implants or environmental sensors deployed in fragile ecosystems.

The study was published in the journal *Nature Communications*.

---

• Researchers at the National University of Singapore developed self-trained lab-grown muscles.

• These muscles powered OstraBot, a biohybrid swimming robot, to a record speed of 467 millimetres per minute.

• The self-training method involves two connected muscle tissues continuously pulling against each other.

Source: INTERESTING ENGINEERING

Alibaba Launches Accio Work, Expands Agentic AI Offerings

Alibaba's international commerce division has launched Accio Work, a new agentic artificial intelligence platform. The "AI taskforce" is designed to autonomously manage complex business operations for small and medium-sized enterprises.

Credit: ALIBABA INTERNATIONAL

Accio Work offers plug-and-play functionality, deploying cross-functional AI teams without requiring coding or setup. This launch marks Alibaba's continued push into the global agentic AI sector.

Alibaba International Vice President Kuo Zhang stated that Accio Work is a specialised B2B tool, not a generalist platform. He emphasised its distinction from the consumer-driven frenzy around agentic AI in China, sparked by OpenClaw and the "lobster raising" trend.

Zhang highlighted the platform's security measures, explaining that any actions involving financial transactions, payment execution, or access to private files require explicit, granular user permission. This approach aims to mitigate risks associated with high-stakes operations.

This introduction follows the recent release of Wukong, another enterprise-focused agentic AI platform from a different Alibaba division. Wukong coordinates multiple AI agents to perform complex business tasks such as document editing, spreadsheet updates, meeting transcription, and research within a single interface.

Alibaba also announced the separation of its AI businesses from its cloud computing arm. This restructuring led to the formation of the Alibaba Token Hub business group, now led by Chief Executive Eddie Wu.

The establishment of Alibaba Token Hub signifies the company's focus on digital assistants powered by AI models. These models utilise significantly more tokens, which are units of data for generating language, than traditional question-and-answer chatbots.

Zhang further commented on the inherent risks in the global effort to define agentic AI. He believes these risks can only be reduced through controlled, specialised models that maintain a balance between automation and security.

He added that the greatest risk comes from using horizontal, generalist models for specific business tasks. By focusing on specialised B2B agents and integrating AI with human approval layers, Accio Work aims to deliver autonomous workforce benefits without the traditional dangers of unconstrained AI.

---

• Alibaba's international commerce division launched Accio Work, an agentic AI platform for small and medium-sized enterprises.

• Accio Work is a B2B tool, requiring explicit user permission for high-stakes operations like financial transactions or private file access.

• This launch follows the introduction of Wukong, another enterprise-focused agentic AI platform from Alibaba.

Source: REUTERS