Why it matters: The introduction of Manus, touted as the world’s first autonomous AI agent, represents a potential leap in AI capability but faces growing skepticism about its underlying technology and marketing claims, highlighting broader questions about transparency in the rapidly evolving AI industry.
“Revolutionary” AI agent under investigation
Chinese startup Monica has unveiled Manus, an AI agent designed to handle complex tasks without human intervention. The system continues working in the cloud even after users disconnect and reportedly outperforms OpenAI’s DeepResearch on the GAIA benchmark.
Manus employs a multi-agent architecture where specialized sub-agents handle different aspects of tasks, allowing it to interact with websites, track activities, and produce various outputs including PDFs and spreadsheets.
Evidence points to Claude Sonnet dependency
Despite claims of full autonomy, investigations suggest Manus may primarily function as a wrapper around Anthropic’s Claude Sonnet model:
- A leaked runtime directory revealed Manus relies on Claude Sonnet alongside 29 tools and open-source software
- Prompt injection tests showed errors in Manus’s responses consistent with Claude Sonnet’s conventions
- Jailbreaking attempts suggest limited communication between Manus’s sub-agents
Industry calls for greater transparency
Critics accuse Monica of using “hunger marketing” by limiting access to select influencers while providing minimal technical details.
The controversy highlights growing industry demands for greater openness in AI development, especially as Chinese AI capabilities continue advancing.
While early demonstrations impressed users with capabilities like automated podcast editing and website creation, questions remain about whether Manus represents a genuine breakthrough or clever integration of existing technologies.