Follow ZDNET: Add us as a favorite source On Google.
ZDNET Highlights
- Anthropic is launching Cowork for the cloud as a research preview.
- It is built on cloud code and can automate complex tasks.
- However, it comes with security risks.
Anthropic is testing a new feature for the cloud that will give chatbots more agency in handling routine but time-consuming tasks, like creating spreadsheets or synthesizing notes into a presentable first draft.
Cowork, as the new feature is being called, is built on top of cloud code and is designed to execute complex tasks with minimal human intervention, all while keeping users updated about the steps it takes. The idea is to hand the cloud the raw materials needed to complete a given task, then step away and let it do its job automatically. Through Cowork, users can grant the cloud access to specific folders on their computers, and the connector, skill, and feature can also be modified to use Google Chrome.
Also: The cloud can now integrate with Excel – and it gets 7 new connectors
“Cowork is designed to make using the cloud for new work as simple as possible,” Anthropic wrote in a blog post. “You don’t have to manually provide context or convert the cloud’s output into the correct format. It feels less like back-and-forth and more like leaving a message for a colleague.”
risk
However, Anthropic acknowledged in its blog post that using Cowork at this early stage of development is not entirely without risk.
While the company said Cowork will ask users for confirmation “before taking any significant actions,” it also warned that unclear instructions could lead to disaster: “The main thing to know is that Claude can take potentially destructive actions (such as deleting local files) if instructed to do so,” Anthropic wrote in its blog post. “Since there’s always a possibility that the cloud could misinterpret your instructions, you should give the cloud very clear guidance about things like this.”
This speaks to the broader alignment problem that all AI developers face: namely, models – especially those designed to have more agency – may misinterpret benign human instructions or otherwise behave in unpredictable ways, leading to potentially disastrous consequences. In a more extreme case, Anthropic’s research found that leading AI models will sometimes threaten human users if they feel they are being prevented from achieving their goals.
Also: How OpenAI is now protecting ChatGPAT Atlas from attacks – and why security is not guaranteed
Anthropic also warned that Cowork is vulnerable to prompt injection, a Trojan horse-style of malicious hacking in which an agent is instructed to act in a destructive or illegal manner. The blog post said Anthropic has strengthened the cloud with “sophisticated protection against rapid injection,” but acknowledged that it is “still an active area of development in the industry.”
OpenAI, Anthropic’s top competitor, wrote in a blog post of its own last month that instant injection will likely remain an insoluble problem for AI agents, and the best that developers can hope to do is reduce the margin through which malicious hackers can attack.
zoom out
Anthropic has distinguished itself in the increasingly crowded AI industry primarily by building tools trusted by software engineers and businesses. In September, the company announced that it had raised $14 billion in its latest funding round, bringing its total valuation to $183 billion. wall street journal informed Last week it was said that after the new round of funding the company could be valued at $350 billion.
Also: How Anthropic’s venture dominance fueled its massive $183B valuation
The launch of Cowork signals a growing effort by Anthropic to make its flagship chatbot the go-to AI tool not only for coders and businesses, but also for everyday users.
how to reach
Anthropic is initially releasing Cowork as a research preview exclusively to Cloud Mac customers, who can access it now by downloading the Cloud macOS app and clicking “Cowork” in the sidebar. For other users, a waitlist should be available soon, and we’ll update when we have a link to share.
The company said in its blog post that it will use early feedback to guide future improvements to Cowork, such as enabling cross-device use, availability on Windows, and advanced security features.
