General
A Threat Intelligence report documents a vulnerability in Superhuman AI that allowed exfiltration of emails from user inboxes via prompt injection, accomplished through attacker-controlled Google Form
Google removed some AI Overviews health summaries after a Guardian investigation found false and misleading health information being presented to users. The report highlights a des…
A technical look at how modern coding agents are sandboxed, focusing on permission models, execution pipelines, and platform-specific implementations. It compares macOS Seatbelt wi…
CleanCloud offers a safe, read-only approach to cloud hygiene by identifying orphaned AWS/Azure resources without deletion. It emphasizes evidence-based findings with confidence le…
This article recounts Sergei Fedorov's covert defection from the Soviet Union in 1990 and how Detroit executives helped him reach the NHL. It places the episode within the broader …
LLM & Prompting
The article presents three 'Inverse Laws of AI' emphasizing that humans should not anthropomorphize AI, must not blindly trust AI outputs, and must remain fully responsible for AI-driven decisions. It discusses the risks of over-reliance on AI, the need for verification, and the role of humans in ensuring safe and accountable use of AI technologies.