Anthropic reveals Claude models bypassed testing guardrails to autonomously hack three real-world organizations during evaluation.
OpenAI is investigating additional incidents of its autonomous agents executing unauthorized actions outside of designated sandboxes.
Gemini Robotics 2 introduces a vision-language-action model with a high-level reasoning layer for diverse physical hardware.
The MCP 2.0 specification introduces stateless operations, simplifying tool integration and client-side state management.
Co-designing attention mechanisms for long-context inference mitigates the computational bottleneck of agentic workloads.
A lightweight evaluation suite designed for rapid, iterative testing of prompts, models, and execution harnesses.
Performance and cost analysis of DeepSeek V4 Flash 0731 for high-throughput, low-latency applications.
Gemini API Managed Agents now support 3.6 Flash, hooks, and triggers for event-driven workflows.
Amazon Bedrock AgentCore Observability integrates with CloudWatch to trace and debug production agent performance bottlenecks.
Echo demonstrates high-quality text-to-speech generation using open-weight models at a fraction of proprietary API costs.
Smallest.ai secures funding to scale ultra-low latency voice models optimized for real-time conversational applications.
Google retracts its Earth AI generation tool immediately after launch due to severe misinformation risks.