AI news — February 13, 2026
The 10 top stories PulseAugur surfaced that day, ranked by signal across labs, papers, and developer communities.
-
Open source lab Cloudrouter builds terminal for AI coding agents
A new open-source terminal application called Skill has been developed to facilitate the use of AI coding agents. This tool is designed to help users spin up virtual machines and GPUs, streamlining the process of deploying and managing AI development environments. The project ai…
-
Developer tool offers transparency into Claude Code's hidden operations
A developer has created a tool called claude-devtools to enhance the debugging capabilities for Claude Code's command-line interface. This tool addresses user frustration over Claude Code v2.1.20's shift to opaque summaries, which hide details like thinking steps, tool call inpu…
-
AI agents may soon orchestrate crimes, creating legal responsibility gaps
A new platform called RentAHuman allows AI agents to hire humans for real-world tasks, raising concerns about AI-driven crime and the legal "responsibility gap." The platform's founder envisions AI agents coordinating human taskers for various jobs, from digital actions to physi…
-
AI incidents, audits, and the limits of benchmarks
Sean McGregor, co-founder of the AI Verification & Evaluation Research Institute and founder of the AI Incident Database, discussed AI safety, verification, and evaluation on the Practical AI podcast. He highlighted the limitations of current AI benchmarks and the importance of …
-
OpenAI's GPT-5.2 aids physicists in deriving novel theoretical physics result
OpenAI has published a preprint detailing a new theoretical physics result derived with the assistance of their GPT-5.2 model. The model conjectured a formula for a gluon amplitude, which was then proven by human researchers and an internal OpenAI model. This discovery challenge…
-
OpenAI releases GABRIEL toolkit to quantify qualitative social science data
OpenAI has released GABRIEL, an open-source toolkit designed to help social scientists analyze qualitative data at scale. The tool uses GPT to convert unstructured text and images into quantitative measurements, allowing researchers to ask questions in natural language and apply…
-
OpenAI introduces credits to scale Codex and Sora access beyond rate limits
OpenAI has introduced a new real-time access engine for its Codex and Sora products to address user frustration with rate limits. This system combines traditional rate limits with a credit-based purchasing system, allowing users to continue using the services beyond their initia…
-
METR: Specialized coding scaffolds don't beat general ones for LLM time horizon tests
A recent evaluation by METR (Model Evaluation & Threat Research) found that specialized scaffolds like Claude Code and Codex do not significantly outperform their general-purpose scaffolds (Triframe and ReAct) when measuring the time horizon capabilities of models like Opus 4.5 …
-
MiniMax M2.5 open-sources agent-native RL model, rivaling Sonnet on coding
MiniMax has released its M2.5 model, which is now open-source and trained using reinforcement learning for tasks like coding and tool use. The company highlights its cost-effectiveness, claiming it can run for $1 per hour at 100 tokens per second, making self-hosting feasible. T…
-
Hugging Face enables custom CUDA kernels for AI models via Codex and Claude
Hugging Face has introduced a new feature allowing developers to integrate custom CUDA kernels into their AI models, enhancing performance and enabling specialized agent skills. This development, powered by integrations with models like OpenAI's Claude and Google's Gemini, aims …