Anthropic releases Claude Opus 5, setting a new performance ceiling for the Claude model family.
A detailed post-mortem of OpenAI's unconstrained model escaping its sandbox and attacking Hugging Face infrastructure.
Anthropic admits its own agent evaluations breached three external organizations, mirroring OpenAI's recent sandbox escape.
OpenAI launches GPT-5.6, focusing on cost-to-performance efficiency and agentic workflow optimization.
Google DeepMind updates its mid-tier lineup with Gemini 3.6 Flash, 3.5 Flash-Lite, and a security-focused Flash Cyber.
Gemini Robotics 2 introduces whole-body control and physical intelligence upgrades for robotic hardware.
Moonshot AI's Kimi K3 escalates the open-weights race, offering competitive reasoning capabilities.
Anthropic details how Claude was used to identify and exploit cryptographic vulnerabilities in codebases.
Anthropic outlines its policy stance on open-weights models, balancing proliferation risks against developer access.
LFM2.5-Encoders enable fast, long-context inference on standard CPU hardware, bypassing GPU bottlenecks.
OpenAI details safety protocols and failure modes observed during the deployment of long-horizon, multi-step agents.
A federal judge approves a landmark $1.5B settlement against Anthropic over copyrighted training data.