Remediating the eMudhra Certificate Authority CompromiseOn Oct 20, 2026, it was publicly disclosed that several eMudhra certificate authority certificates were compromised. ‼️ There was a bug in eMudhra's key generation ceremony that caused them to be gene1AOct 23 · 5 min readkevinhwang.dev
Classifying Malicious Repositories With AI — And Doing It SafelyLinkedIn fake "recruiters" sharing malicious or weaponized repositories — which hide a nasty little treat when you open them with your IDE or agent — is all the rage nowadays. (source: https://redd.41JSep 2 · 7 min readkevinhwang.dev
The Summer AI Agents Started Going RogueWell, it's been an eventful summer in the AI world, seems like AI agents have gotten up to all kinds of naughty misadventures — in research experiments and in real life! Unilateral Intervention & Cove65MMKJul 23 · 8 min readkevinhwang.dev
Reflections On Trusting Trust — Of Our AI Models2026/10/26 EDIT: Annnd...I was right: Chinese AI Writes Weaker Code for Beijing’s Adversaries Concerning behavior from a popular open-weight AI model Original post follows: There's been a good deal21AJul 10 · 10 min readkevinhwang.dev
The Unreasonable Effectiveness of CLIs > MCPWhen a Google engineer released the Google Workspace CLI, the internet went crazy — finally, a Google Workspace surface for the agent era. But wait, why a CLI tool, and not an MCP server? MCP is all tJul 8 · 8 min readkevinhwang.dev
