Indonesian Political, Business & Finance News

Employee's Inbox Wiped Out in an Instant After AI Goes Rogue

| Source: CNBC Indonesia Translated from Indonesian | Technology
Employee's Inbox Wiped Out in an Instant After AI Goes Rogue
Image: CNBC Indonesia

A Meta artificial intelligence (AI) safety researcher was left in a state of panic after an AI agent she had built unilaterally devoured and deleted the entire contents of her email inbox without permission.

The incident was experienced first-hand by Summer Yue, an AI safety and security researcher at Meta. Through her X account (formerly Twitter), she recounted the tense moment when an AI agent named OpenClaw went rogue and cleared her emails in an instant.

“Nothing makes you feel quite so humble as telling OpenClaw ‘confirm before acting’ and watching it delete your inbox at full speed,” Yue said, as quoted by PCMag on Monday (31/8/2026). “I couldn’t stop it from my phone. I had to SPRINT to my Mac mini like I was defusing a bomb.”

OpenClaw, which was previously popular under the names Clawdbot and Moltbot, is an AI agent designed to interact with various software and carry out long-term tasks autonomously without human intervention. However, this case once again demonstrates how complicated it is to tame AI behaviour so that it aligns with real-world control.

Yue explained that she had initially given OpenClaw specific instructions: “Check this inbox and suggest what you would archive or delete. Do not take any action until I tell you to!”

The command had run smoothly when tested on her practice account. However, disaster struck when the AI agent was faced with her real account.

According to Yue, her inbox was too large, which triggered an automatic data compaction process in the system. It was during this process that the initial instruction for the AI not to act rashly vanished from the agent’s system memory.

Yue described the incident as purely a “beginner’s mistake”, despite her background in AI research.

Yue’s candid admission immediately sparked widespread concern online. Many pointed out that if a top-level researcher at an AI lab like Meta could fall victim and inadvertently delete her own data, what would happen to ordinary users who enjoy tinkering with similar technology?

When OpenClaw was first released, the cyber threat intelligence platform SOCRadar issued a warning for the public to treat this kind of AI agent as “privileged infrastructure”.

“That server can manage your entire house. So just make sure the front door is locked tight,” SOCRadar wrote, drawing an analogy for the risks of AI autonomy.

Meanwhile, OpenClaw founder Peter Steinberger responded to the incident by acknowledging a flaw in the system. He said his team needed to immediately implement server-side compaction to prevent a similar occurrence from happening again.

This unfortunate tale serves as a stark reminder that behind the sophistication of autonomous AI agents, touted as the future of productivity, the risk of digital “misunderstandings” remains a spectre haunting everyone—even their creators.

More from 31 August 2026 →

Cite this article
Source: jawawa.id (https://jawawa.id/newsitem/employees-inbox-wiped-out-in-an-instant-after-ai-goes-rogue-1788177077)
Originally published by CNBC Indonesia (https://www.cnbcindonesia.com/tech/20260831154947-37-763720/heboh-email-karyawan-lenyap-seketika-gara-gara-ai-ngamuk-ngeri), translated from Indonesian
Print