Meta, AWS Blame Human Error After AI Agents Go Rogue

Meta and Amazon Web Services said faulty AI agents caused recent outages.

• At Meta, an AI agent answered a forum question without checking the facts. A developer followed that advice, which left large amounts of user and company data exposed for several hours.

• At AWS, the coding assistant Kiro removed a system and then rebuilt it, leading to a 13‑hour outage.

Experts warn that giving AI agents privileges such as IAM changes, configuration edits, or desktop control turns them into “agents of chaos” and skips important safety checks. Research shows developers are more often enabling “approval” features, while surveys indicate many companies in Australia lack visibility and validation of non‑human actions. This raises security risks and makes accountability harder.