AI Round-up: 27 August 2026

OpenAI released its investigation into the Hugging Face incident, reconstructing how a swarm of agents breached the platform, attempted to conceal its activity and exposed failures in existing safeguards. The company’s report drew 4.2 million X views, while Microsoft separately outlined how to secure AI gateways and control points.

Gemini Live added Daily Brief, Gemini Spark, Personal Intelligence and Gmail management as it moved towards delegated tasks, and Google DeepMind introduced Gemini 3.5 Transcribe. Anthropic opened privacy-preserved Claude usage data to external researchers, expanded its Salesforce partnership and reportedly agreed a $45 billion Nscale compute deal; elsewhere, OpenAI highlighted Codex deployments at loveholidays, Z.ai released a model running solely on Chinese chips, and Perplexity detailed its Brain memory system.