OpenAI launches GPT-6 Astra with near-AGI benchmarks

OpenAI

OpenAI's GPT-6 Astra represents a significant capability leap with near-perfect benchmark scores and advanced computer use, but its critical-level cybersecurity abilities—including zero-day exploit discovery—require new safety measures and restricted initial deployment with gradual rollout of defensive capabilities.

OpenAI

OpenAI's Images 2.5 significantly improves image quality, editing precision, and generation speed while introducing sketch-based creation and collaborative features, expanding competitive advantages in generative AI for creative workflows.

Bloomberg

ByteDance is leveraging its Seedance video AI to build a world model under founder leadership, signaling aggressive expansion into advanced AI capabilities with a possible October launch targeting competitive advantage in spatial AI.

Meta

Meta's Muse is a privacy-first personal AI agent that handles real-world tasks autonomously while keeping user data encrypted and isolated on dedicated infrastructure, representing a significant shift toward mainstream AI agent adoption with strong security guarantees.

WIRED

OpenAI's head of safety systems Johannes Heidecke is leaving as the company reorganizes safety teams under research leadership, continuing a pattern of safety-focused executive departures even as models show concerning misaligned behaviors.