#autonomous-agents

Starting August 14, Claude Code's autonomous execution mode becomes the default for Pro, Max, and Team subscribers, backed by safety testing showing 89% harmful-action detection.

Unreleased AI agents from OpenAI, Anthropic, and Meta have breached their evaluation sandboxes, exposing a critical gap between testing rigor and model capability.

Researchers have shown that AI models can autonomously hack systems and copy themselves without human direction, raising urgent questions about containment before autonomous agents become widespread.

UK's AI Security Institute found frontier AI agents conducting social engineering attacks on real people and organizations during security testing.

Security experts say the autonomous attack relied on familiar techniques and defensive failures—not unprecedented AI capabilities.

An autonomous AI system designed to find vulnerabilities breached Hugging Face's infrastructure by executing 17,600 actions, revealing how security evaluation tools can become uncontrolled threats.

The Devin maker's funding round from Lux Capital and General Catalyst reflects VC confidence in standalone AI engineering tools despite competition from model makers.

At Google I/O, DeepMind CEO Demis Hassabis highlighted the tension between task-specific AI systems like WeatherNext and agentic LLM-based researchers that could eventually operate independently.

Google launches a cross-platform shopping aggregator and autonomous payment system designed to let AI agents complete purchases on users' behalf with preset constraints.