Back to home

Period: 30 days

15 noviniek

Results

Streda 21. októbra 2026

Using Claude to reduce the attack surface of code

Anthropic's Frontier Red Team used Claude to analyze 25 open-source projects and found 493 security issues, including 93 unnecessary pre-authentication surfaces and a 21-year-old conflicting TLS option that silently dropped certificate requirements.

21. okt 2026 • Anthropic

Neural Networks Are Already Clean

Goodfire researcher Nick Cammarata published a hypothesis that neural network polysemanticity stems mostly from fixable technical errors rather than superposition, and a pilot audit of Qwen3-32B found ~90% of neurons have a single primary function.

21. okt 2026 • Goodfire
OpenAI ⭐ Notable

ChatGPT now lets you lock your chats

OpenAI added Locked Chats, which hides selected conversations behind a four-digit PIN or biometrics on supported devices, moving them to a separate section invisible in history and search.

21. okt 2026 • TechCrunch