Cursor released Composer 2 without disclosing its base model; calling its OpenAI-compatible API revealed it is Kimi K2.5. This escalated into a licensing dispute, but a formal commercial agreement with Moonshot AI was subsequently confirmed.
AttnRes to replace Transformer's fixed residual combination with softmax attention in the depth direction. Demonstration with Kimi Linear 48B improved GPQA-Diamond +7.5pt and HumanEval +3.1pt. Training overhead was kept below 4% and inference below 2%.
Bartz v. Anthropic, a class action lawsuit over learning through unauthorized downloading from LibGen, etc., reaches settlement. FSF, as the copyright holder of the GNU FDL works, received the settlement notice and announced its position demanding disclosure of the training data model for LLM learning.
On March 19, 2026, Anthropic took legal action against OpenCode to remove the OAuth integration. On the same day, Python toolchain Astral announced that it was joining OpenAI's Codex team. The formation of AI coding tools was activated in one day.
Claude Code v2.1.80 introduced the research preview feature Channels, which lets Telegram and Discord send live instructions into a Claude Code session through MCP. This article summarizes the bidirectional design and security model.
An AI system that automatically reviews Linux kernel patches. Detected 53.6% of known bugs that passed human review. Rust implementation supports both Gemini and Claude.
NVIDIA's NemoClaw secures OpenClaw agents with a four-layer sandbox, while Stripe's Machine Payments Protocol lets agents make payments without handing over private keys. The open question is how to charge safely from inside the sandbox.
A buffer overflow was discovered in the LINEMODE SLC handler of GNU Inetutils telnetd. No authentication required - root privileges can be gained just by connecting to port 23. All versions (~2.7) are affected and no patch has been released.
CLI-Anything, released by HKUDS at the University of Hong Kong, automatically generates a CLI harness from GUI software source code so AI agents can drive it directly. It passed all 1,720 tests across 16 apps including GIMP, Blender, and LibreOffice.
Cisco Secure FMC's unauthenticated RCE flaw CVE-2026-20131 (CVSS 10.0) was added to CISA's KEV catalog after Interlock ransomware had been abusing it for 36 days before Cisco's public disclosure. Amazon Threat Intelligence later dissected the toolkit in detail.
H Company's Holotron-12B uses a memory-efficient new design to lift PC-operation AI throughput to 8,900 tokens per second. Unsloth has released the beta of 'Studio,' a browser tool for no-code model fine-tuning.
