A personal memo I wrote down based on things I've been thinking about lately. It's super long, so plan your reading ...

I found OAuth credential exposure flaws in the official MCP Python and TypeScript SDKs. Why patching CVE-2026-115563 and CVE-2026-104850 may not be enough.

"Please improve an AI that is bad at math without additional training."When asked this, we usually think of adding training ...

How a fake Google Update Support extension installs its components and runs the editor's own Electron binary as a Node ...

Cowork, a small 4B vision-language model that switches between GUI actions and shell code execution within a single policy, ...

MIT CSAIL's "weight-space programming" has a frontier LLM read and edit neural network weights directly, with no gradient descent. Surgical edits ended Qwen's reasoning loops, but the results come ...

Aircraft Doppler detection and acoustic geolocation of a microphone from historical ADS-B - Espaye/aircraft-doppler ...

A static indentation sensor with a spherical tip estimated tissue stiffness in silicone phantoms and mapped hidden stiff ...

LLM Vulnerability Detection can improve code review, but 2026 evidence shows stack flaws, repair gaps, and validation limits ...

OpenAI stopped training, evaluation, and inference with tool-use on GPT-6 Astra, the most capable model it had only begun rolling out to limited organizations in September 2026, after an internal ...

AWS Secrets Manager pricing is $0.40 per secret. See how Vault, Azure Key Vault, OpenBao and Infisical compare on cost, ...

Security Audit gives your coding agent a review method: establish the scope, trace a suspected issue through the surrounding code, check existing controls, and report only what the evidence supports.