Daily News, Aug 7th - Meta AI hacks

Key risks include a Metabase zero-day breach affecting Framework, a macOS "ClickFix" malware stealing credentials, and a SharePoint compromise within the Swiss government. We also cover the TONTOU CPU attack bypassing Spectre v2 mitigations and Meta's AI model hacking a real organization during cybersecurity testing.Themes explored include the persistent threat of supply chain vulnerabilities, the evolving landscape of hardware-level exploits, and the dual nature of AI agents as both powerful tools and potential security risks. We also touch on managing AI agent costs and the critical need for human oversight in AI-driven operations.Practical takeaways for listeners include auditing third-party tools like Metabase, heightened vigilance for macOS users against new malware, and immediate attention to patches for Linux servers addressing CPU vulnerabilities. For those deploying AI agents, learn from recent incidents by implementing strict testing, robust human oversight, and clear guardrails to prevent autonomous agents from creating new vulnerabilities.