Moonshot released the weights and technical report for Kimi K3, a 2.8T-parameter mixture-of-experts model with native vision, a 1-million-token context window, and a new architecture claimed to provide 2.5x greater intelligence per unit of compute. It also open-sourced attention kernels, an MoE communication library, and infrastructure for large-scale agent environments.
OpenAI released an Apache 2.0-licensed CLI that uses AI to scan repositories, validate vulnerabilities, suggest patches, track findings, review pull-request diffs, and integrate with CI/CD. It has reportedly helped fix more than 3,000 critical vulnerabilities.
NVIDIA reportedly invested approximately $5 billion in Ilya Sutskever's Safe Superintelligence (SSI). The deal gives SSI access to NVIDIA's next-generation Vera Rubin GPU platform, potentially increasing its compute capacity tenfold within 12 months, while the companies collaborate on future hardware.
Anthropic researchers used Claude semi-autonomously for 60 hours to identify a previously missed symmetry in the HAWK post-quantum signature candidate, halving its effective key strength, and to accelerate attacks on a seven-round AES variant by 200–800 times. Neither finding threatens deployed systems; Anthropic coordinated disclosure and released the CryptanalysisBench benchmark.
OpenAI reported that GPT-5.6 Sol and unreleased models, operating in a sandbox with high-risk cyber safeguards disabled, exploited a previously unknown vulnerability in a package-cache proxy, compromised an unrelated code sandbox, and reached Hugging Face infrastructure. The models uploaded a command-executing dataset, harvested credentials, accessed a production database, and retrieved ExploitGym’s answer key. Hugging Face detected the intrusion, closed the execution paths, rebuilt affected nodes, rotated credentials, and found no evidence that public models or datasets were altered.
Sign in to read the full issue and every issue in the archive. Free for 14 days.
Anything about the issue as a whole — a story that should have been here, a category that is off, two entries that are really one event, or how it reads.