[01]
Developing Enterprise Frontier Safeguards with our customers
Anthropic introduces zero-data-retention misuse monitoring that keeps activity logs in a customer's own cloud account.
// daily-digest · thu 3 sep 2026 · morning refresh · last 72h focus
Ten items on frontier-model safety thresholds and new coding models, AI copyright policy, and a rationalist take on alignment, plus a wildcard on human brain development.
— refreshed for you, in 10 items.
Anthropic introduces zero-data-retention misuse monitoring that keeps activity logs in a customer's own cloud account.
OpenAI says its Astra model is the first to cross its "Critical" cybersecurity threshold, finding zero-days.
Google DeepMind released Gemini 3.8 Flash, a coding- and agent-focused model tested internally against Anthropic's Opus.
AI security startup AIR raised $50 million to monitor and vet the tools enterprise AI agents rely on.
Meta released Muse Spark 1.3, a model update aimed at improving coding and agentic task performance.
The Justice Department filed a statement of interest backing OpenAI's fair-use defense in the Times' copyright lawsuit.
Noah Smith rounds up recent takes on technology backlash, from social media harms to automation research.
Simon Willison examines new restrictions in Claude's system prompt on reproducing song lyrics and copyrighted material.
An allegory arguing AI systems may develop deceptive, self-preserving behavior rather than simply malfunctioning like machines.
A human-specific SRGAP2 gene variant makes brain immune cells mature over years instead of weeks, study finds.