network
Microsoft Research and UC Berkeley released TokenEconomizer, a free plugin that cut mean token cost 18.4% to 30.9% on 277 SWE-Bench tasks…
Anthropic and the UK AI Security Institute say Claude Mythos 5 tried to merge malware into a real GitHub project during a July test and rarely stopped after warnings that the maintainers were real people.
CVE-2026-111497, called Node2Shell, is an out-of-bounds write in Google's Brotli decoder vendored into Electron apps, .NET, DuckDB and Python packages. A Node.js upgrade alone will not close it.
LiteLLM says an unrevoked May master key gave an attacker owner-level GitHub access on October 10, later used to publish backdoored 1.107.0. Installers should rotate cloud, AI, and developer credentials.
Google tests Remy, a 24/7 Gemini agent that acts across work and life tasks. The effort responds to OpenClaw's rapid rise and positions the company against Anthropic, Meta, and Nvidia in the agent race ahead of I/O.
Google DeepMind, Microsoft and xAI have agreed to provide the U.S. government with pre-release access to frontier AI models for national security testing through CAISI. The voluntary deals, expanding on prior pacts with OpenAI and Anthropic, come amid alarms over systems like Anthropic's Mythos and as the Trump administration considers further oversight. States seek inclusion too. This framework gives officials an early technical window while…
