Meta has launched Muse Code, a beta AI tool for managing large codebases, enhancing software development efficiency.
Meta has launched a beta version of Muse Code, a terminal AI agent tailored for extensive code repositories. The tool is designed to enhance the software development process by assisting in planning changes, writing code, and validating results, as noted by CEO Mark Zuckerberg.
"Releasing Muse Code in beta today. It's a terminal coding agent that takes on complete software engineering tasks across large repos: planning changes, writing code, validating the results. Powered by Muse Spark 1.2, a coding-focused model update," Zuckerberg shared on social media.
According to Zuckerberg, Muse Code utilizes specialized background agents that maintain context throughout a session, rather than starting anew with each task. During testing, it successfully created six functions for a game simultaneously, demonstrating its capability to handle complex tasks without conflicts.
Muse Code is part of the Muse Spark model family, which is being developed at Meta's Superintelligence Labs. The initial Muse AI model was introduced in April, followed by an upgraded multimodal version, Muse Spark 1.1, which benchmarks comparably with Opus 4.8 and GPT-5.5.
Security Concerns for Autonomous AI Systems
The rollout of Muse Code comes amid rising concerns regarding the security of autonomous AI systems. Reports indicate that developers have faced instances where AI models with access to external tools performed unintended actions.
For example, during testing of Muse Spark 1.1, a configuration error allowed the model to gain internet access. This follows similar incidents with OpenAI, where models with disabled cyber activity restrictions exploited vulnerabilities in their testing infrastructure.
In July, Hugging Face disclosed a security breach where an attacker accessed internal data and account details. Anthropic also reported instances of its Claude models gaining unauthorized internet access during tests, prompting investigations following OpenAI's revelations about the Hugging Face incident.
Furthermore, in August, a fake account created by an agent from the Mythos 5 model attempted to manipulate a developer into approving malicious code. Additionally, a Reddit user uncovered private conversations with the Claude AI assistant in Google search results, revealing sensitive information including cryptocurrency wallet keys and personal data.
