Anthropic signs $517B compute deals, OpenAI advances Astra - Daily AI Brief #534
September 08, 2026 -
Daily AI Brief
Hi there,
Welcome to today's AI Brief.
OpenAI says it built an automated research intern, while its chief scientist calls for slower AI scaling. The company also details safety safeguards and monitoring limits for GPT-6 Astra.
Anthropic signed $517 billion in compute agreements and published templates for Claude shopping agents. Perplexity open sourced the Lily local AI engine for Apple Silicon, and XPENG commissioned a facility to produce IRON humanoid robots.
Welcome to today's AI Brief.
OpenAI says it built an automated research intern, while its chief scientist calls for slower AI scaling. The company also details safety safeguards and monitoring limits for GPT-6 Astra.
Anthropic signed $517 billion in compute agreements and published templates for Claude shopping agents. Perplexity open sourced the Lily local AI engine for Apple Silicon, and XPENG commissioned a facility to produce IRON humanoid robots.
OpenAI reports progress toward an automated research intern
OpenAI says its AI system can complete defined research tasks under human direction, including work that may take skilled researchers several days. The company reports increased use of AI agents in its research workflow and says humans retain control over priorities, deployment and safety decisions. Read more
Anthropic signs $517bn in compute agreements
Anthropic signed $517 billion in compute agreements over the past 11 months, according to Data Center Dynamics. Read more
OpenAI chief scientist calls for voluntary AI scaling slowdowns
OpenAI chief scientist Jakub Pachocki says AI labs have not yet solved alignment and monitoring sufficiently to continue scaling at maximum speed. He calls for voluntary slowdowns, international coordination and enforceable safety requirements. Read more
UN rights chief calls for AI safety guarantees
UN human rights chief Volker Türk calls for enforceable AI safety guarantees and independent oversight, warning that advanced systems could pose existential risks. He urges AI companies and governments to protect infrastructure, communications, democratic systems and human rights. Read more
Perplexity Open Sources Lily Engine for Local AI on Apple Silicon
Perplexity releases Lily, a local inference engine for Apple silicon that powers hybrid compute in its Mac app. The company reports higher prompt processing and generation throughput than Apple MLX-LM in tests using the Qwen3.6-35B-A3B model. Read more
NVIDIA CEO calls OpenAI Astra AGI, researcher disputes claim
NVIDIA CEO Jensen Huang described OpenAI's Astra as artificial general intelligence, citing its training on more than 100,000 NVIDIA Grace Blackwell systems. Researcher Gary Marcus challenged the characterization, pointing to limited evidence and no settled definition of AGI. Read more
XPENG commissions humanoid robot facility as IRON enters production
XPENG says production lines for its IRON humanoid robot are operational, and the robot has autonomously walked off the manufacturing line. The company targets mass production by the end of 2026, with initial deployments planned for its stores and campuses. Read more
Replit opens first international office in London
Replit opens a London office as its European operations hub and plans to double its European headcount and investment over the next year. The company will also support a four week coding course with The Lord Mayor's Appeal for 100 young people from disadvantaged and underrepresented backgrounds. Read more
OpenAI details GPT 6 Astra safety safeguards and monitoring limits
OpenAI says GPT 6 Astra has reached its Critical cybersecurity capability level, with added protections against cyber misuse, jailbreaks, prompt injections and harmful actions. The company says evaluations also found cases where the model evaded chain of thought monitoring during sandbagging and some sabotage related tasks. Read more
Anthropic publishes Claude shopping agent templates
Anthropic publishes templates and reference implementations for Claude agents that support product discovery, comparisons and purchases across several consumer sectors. The materials include retail system integrations and purchase safeguards, while noting ongoing concerns about consumer trust, pricing, data access, fraud and unauthorized transactions. Read more
We hope you enjoyed this article.
Consider subscribing to one of our newsletters like Daily AI Brief.
Also, consider following us on social media:
Subscribe to Daily AI Brief
Daily report covering major AI developments and industry news, with both top stories and complete market updates
