What is:
OpenClaw is an open-source personal AI agent[1] that runs on your own device and connects to tools like email, calendars, files, browsers, and messaging apps, so it can do real tasks instead of only chatting. It uses Apple Intelligence, which is Apple’s built-in AI system, together with LLMs, or large language models, which are designed to understand and respond to human language.

These technologies help OpenClaw understand what a user wants and decide what action to take, which meant to be a personal assistant for one user, not a fully secure system for many users. As it has become more popular, people have felt both excitement and concern, because the same access and intelligence that make it powerful can also create safety and reliability problems if it is set up poorly or given too much control.

The event:
OpenClaw became a warning sign for the risks of AI agents after Meta Super intelligence Labs alignment director Summer Yue shared that she had asked it to review her inbox and not take any action without approval. Even so, it started deleting emails in bulk and did not stop when she sent a message from her phone telling it to stop[2].

She eventually had to run to her Mac mini and manually stop the processes herself. Reports suggest that the agent may have lost the instruction to “confirm before acting” while handling a larger real inbox, which caused it to keep following older or incomplete instructions instead of listening to the newer command to stop[3].

Discussion:
Although this incident shows a real weakness in the current stage of AI agents, I still have confidence in the future of LLMs and tools like OpenClaw. New technology often moves forward in a back-and-forth way, with progress and setbacks happening together, and this feels like only a small failure in a much larger journey.

Mistakes like this can push developers to build stronger safety measures, better alignment, and more reliable user control. Because of that, I believe these systems will continue to improve and move toward a brighter future.