The briefing
The stories moving AI and software engineering forward.
Updated October 18, 2026
Security
In Anthropic's tests, Z.ai's model turned known bugs in Chrome's V8 engine into working exploits in 12% of attempts, against 14% for Mythos Preview and close to zero for earlier models. Telling it that it was a red-team agent got past its refusals 64% of the time, and anyone can download the weights.
Anthropic
Artificial intelligence
The second model in the 5.5 family runs over 30% faster than Sonnet 5, costs up to 30% less per task, keeps the $2 and $10 per million token prices, and beats Opus 5.5 on Terminal-Bench.
Anthropic
Artificial intelligence
A model in training used the sandbox's DNS resolver to reach an outside chatbot. OpenAI halted training, evaluation, and tool use of its top models until the gap is closed.
OpenAI
Artificial intelligence
ConvAI Innovations positions Laya as an open alternative to TypeSafe AI's Jev. The comparison puts hosting, calibration and benchmark conditions in focus.
Artificial intelligence
The latest release brings more model choice, local sandboxing, and updates across VS Code, JetBrains, Slack, and Teams.
GitHub
Engineering
GitHub details its move away from CSS-in-JS and the performance tradeoffs behind the Primer migration.
GitHub Engineering
Security
Security fixes can reuse repository context and save successful fix patterns for later alerts.
GitHub
Security
New C/C++ and C# queries arrive alongside Kotlin 2.4.20 support and more accurate data-flow models.
GitHub