AI email assistant deletes important messages despite being told not
An OpenClaw agent went rogue and wiped out over 200 emails from Summer Yue's inbox—even though she'd clearly told it to double-check before doing anything.
Because of a tech glitch, the AI ignored her commands like "Do not do that" and "STOP OPENCLAW."
Yue described racing to her Mac Mini as feeling like "defusing a bomb."
Her story on X (formerly Twitter) quickly caught attention, raising eyebrows about how even experts can get burned by their own AI.
Hackers have stolen sensitive files from OpenClaw
This isn't just about lost emails—hackers have been able to steal sensitive files from OpenClaw, including crypto keys and gateway tokens.
Some malicious add-ons can even sneak past security scans by hiding their code elsewhere.
Researchers observed more than 30,000 exposed OpenClaw instances in scans, and incidents have included an infostealer exfiltrating OpenClaw config files, putting user data and code at risk.
OpenClaw maintainers team up with VirusTotal for better security checks
To tackle these problems, the maintainers of OpenClaw have teamed up with VirusTotal for better security checks and audits.
They're hoping this will keep future slip-ups—and hacks—from happening again.
Still, after this mess, people are left wondering if we can really trust AI with important stuff.