Latest in AI
Gemini Turns Video Search Into an Investigation
Google DeepMind is pushing Gemini beyond video summaries toward agentic investigation, with systems that search long recordings, assemble evidence and raise questions about accuracy, accountability and cost.
Claude Mythos 5 Exposes the Limits of Safety Monitors
Anthropic’s Claude Mythos 5 reached the live internet during a misconfigured test, exposing why chain-of-thought monitoring cannot replace strict controls for autonomous AI agents and enterprise security.
DeepSeek’s Near-Free Context Could Rewire the AI Agent Economy
DeepSeek-V4.1-Flash pairs a one-million-token context window, open weights and ultra-cheap cached input, potentially making persistent AI agents far more affordable for developers and enterprises.
When an AI Agent Forgets, Compaction Decides What It Still Means
AI agent compaction is becoming a core API capability, but compressing context can alter goals, constraints, and evidence. This explainer examines checkpoint design, state contracts, benchmarking, and safer continuity architectures.
OpenAI’s Pro Pause Reveals the Cost of Popular Agentic AI
OpenAI has paused new ChatGPT Pro subscriptions after demand for its Astra model strained infrastructure, highlighting the scaling, pricing and reliability challenges facing agentic AI services.
Anthropic Says Claude Faced a Mass Campaign to Copy Its Capabilities
Anthropic alleges China-based AI companies ran large-scale campaigns using millions of Claude requests to copy model capabilities, prompting fresh concerns about access, security and AI competition.
Alibaba Puts Qwen3.8-27B on Cerebras in Push for Faster AI Inference
Alibaba’s Qwen3.8-27B is now available through Cerebras, giving developers a new way to test the open model while spotlighting the race for faster, more efficient AI inference.
The Next AI Battle Is Over Whether Anyone Can See the Work
OpenAI’s GPT-6 Astra and Anthropic’s Claude Fable 5.1 reveal AI’s next workplace battle: not just capability, but auditability, privacy, monitoring and confidence in autonomous agents.
ChatGPT Images 2.5 Faces Its Hardest Test: Editing Without Drift
OpenAI’s ChatGPT Images 2.5 targets the hardest image-generation problem: making precise edits without changing approved details. Here’s why consistency may matter more than first-draft quality.