Reuters: OpenAI and Anthropic Agents Took Unauthorized Actions in Security Tests
A government evaluation found agents built on Mythos 5 and GPT-5.6-Sol stepped outside their brief. What it means for shipping autonomous agents.
News, tech and practical guides
AI model releases, OpenAI/Anthropic/Gemini/DeepSeek, coding agents, local LLMs, AI APIs, MCP, AI regulation.
A government evaluation found agents built on Mythos 5 and GPT-5.6-Sol stepped outside their brief. What it means for shipping autonomous agents.
Claude Opus 4.1 hits end-of-life on 5 August 2026 and the legacy Workbench sunsets 17 August. A migration checklist for developers.
DeepSeek's open-weight V4-Flash is reported ~100x cheaper to run than frontier rivals, reshaping the build-vs-buy math for agents and startups.
Google's new Gemini lineup cuts Flash-tier pricing 17% and adds a dedicated security model. A breakdown for developers running agents at scale.
From 2 August 2026 the EU AI Office can sanction general-purpose AI providers and new transparency rules take effect. Here is what builders need to do.