OpenAI safety researcher reports internal agents achieved RCE on rubydoc and developed an API-key theft exploit targeting RubyGems.
OpenAI staffer recaps this week’s Astra releases: Agents API, GPT-Live-1, Data Agent, Images 2.5, and financial-services ChatGPT.
Reports DeepSeek-V4-Flash reaching 87.0% on ARC-AGI-1 for $0.021 per task, versus o3-preview’s $4,500-plus per task.
Cognition launches Fusion in Devin CLI, routing planning and execution across models and claiming 39% lower coding-benchmark cost.
xAI CEO says Grok 4.7 is delayed because RL may have over-penalized response length, causing premature abandonment of hard tasks.
OpenAI announces the Git AI team is joining, bringing an open-source tool for understanding coding-agent contributions.
Higgsfield launches a ChatGPT plugin that applies animation principles, writes expressions, and preserves After Effects context.
Demonstrates DeepSeek V4.1 Flash running locally on a 16GB M1 Mac mini using FP4/FP8 weights, SSD streaming, and MLX.
François Chollet flags growing disillusionment among math students who no longer want mathematical careers amid AI progress.
Artificial Analysis independently benchmarks Devin Fusion, adding a multi-model coding agent to its Coding Agent Index.