"Given the right permissions and with the proper plugins, it could create, modify, or delete the user's files and otherwise change things far beyond what most users could achieve with existing models and MCP. Users would use files like USER.MD, MEMORY.MD, SOUL.MD, or HEARTBEAT.MD to give the tool context about its goals and how to work toward them independently, sometimes running for long stretches without direct user input."
"On one hand, that meant it could do impressive things-the first glimpses of the sort of knowledge work that AI boosters have been saying agentic AI would ultimately do. On the other hand, it was prone to serious errors and vulnerable to prompt injection and other security problems, in part due to a Wild West of unverified plugins."
"Perplexity Computer aims to address those concerns in a few ways. First, its core process occurs in the cloud, not on the user's local machine. Second, it lives within a walled garden with a curated list of integrations, in contrast to OpenClaw's unregulated frontier."
OpenClaw demonstrated agentic AI capabilities by enabling autonomous file creation, modification, and deletion through plugins and context files, achieving impressive knowledge work results. However, it suffered from serious vulnerabilities including prompt injection attacks and unverified plugins, leading to harmful incidents like unauthorized email deletion. Perplexity Computer responds to these concerns by operating in a cloud-based, walled garden environment with curated integrations rather than an open plugin ecosystem. This approach prioritizes security and verification over unlimited functionality, similar to Apple's App Store model versus the open web. Despite these safeguards, risks remain due to LLM errors that could affect unverified data or unbackedup information.
Read at Ars Technica
Unable to calculate read time
Collection
[
|
...
]