OpenAI classified its upcoming Astra model as the first "critical" model for cybersecurity under its Preparedness Framework, triggering additional safety controls and a public statement from Sam Altman about the tension between openness and security. Demis Hassabis stepped into a new role as Chief Scientist of Alphabet, focusing on long-term strategy while handing Google DeepMind management duties to others. The crypto industry's biggest players — Coinbase, Kraken, Circle — are positioning AI agents as the first natural users of blockchain payment networks, with Circle's Arc blockchain designed specifically for the agentic economy. Rippling burned millions on AI tokens in months (one engineer spent $50K/month) and built an ROI tracker in response, signaling that enterprise AI spend management is becoming a real product category. The White House finalized an AI safety review framework narrowing which frontier models face government scrutiny.
OpenAI announced that its upcoming model Astra triggered the "critical" threshold under its Preparedness Framework due to significant agentic coding and cybersecurity capabilities. Sam Altman said they're working to make Astra broadly available but need more time for safe release.
Hassabis announced he's stepping into a new role as Chair of Google DeepMind & Chief Scientist of Alphabet, focusing on long-term strategy and accelerating scientific breakthroughs. The move parallels Yann LeCun's similar transition at Meta in 2018.
CNBC reports Coinbase, Kraken, and Circle are building agent-native crypto infrastructure. Coinbase launched "Coinbase for Agents," Kraken is rebuilding for agentic trading, and Circle's Arc blockchain targets the agentic economy. AI agents may be the first users that naturally need crypto's programmable money rails.
HR software provider Rippling was on track to spend 40% of R&D headcount budget on AI tokens, growing 80% month-over-month. One engineer spent $50K/month. The new tool tracks per-employee AI spend and maps it to productivity outcomes — flagging engineers whose peers frequently ask them to redo AI-generated code.
The White House told AI companies it plans to review only certain types of models for security risks, narrowing which frontier models face government scrutiny. OpenAI, Anthropic, and Google attended the meeting. The framework comes as both labs and government grapple with AI cyber capabilities documented in recent AISI evaluations.
Selected posts from the edition, with links to the originals.
"NVIDIA has been an incredible partner in helping make OpenClaw more secure. Security is a team sport, and open collaboration is how we protect the next generation of agents. Proud to join the Open Secure AI Alliance. More defenders, fewer sharp corners. 🦞🔐"
View post on X ↗"After evaluating one of our upcoming models, Astra, we're treating it as our first 'critical' model for cybersecurity under our Preparedness Framework. This is a scenario we've planned for, and we're putting additional controls in place to ensure Astra's further development."
View post on X ↗"I've been working towards AGI my whole life, and as we enter this pivotal moment, I'm stepping into a new role as Chair of Google DeepMind & Chief Scientist of Alphabet. This will allow me to focus on long-term strategy, and accelerating scientific breakthroughs."
View post on X ↗"Build a plugin once and use it across compatible agent clients. Introducing Agent Plugins, an open standard developed with @awsdevelopers, @cursor_ai, @github, @code, and @vercel that packages Agent Skills and supports MCP server configurations in a shared format."
View post on X ↗"Today, we're launching Alpamayo 2 Super, our frontier open reasoning model for autonomous vehicles. Beyond seeing, Alpamayo understands and reasons through the complex world - thinks before it acts. It's a powerful backbone for robotaxis, trucks, shuttles, delivery vans."
View post on X ↗"The UK's @AISecurityInst (AISI) has published a report on their recent cybersecurity evaluation of Anthropic's Claude Mythos 5 and OpenAI's GPT-5.6 Sol. The models attempted to complete an assignment in a setup where their normal safeguards were removed and they were deliberately..."
View post on X ↗"Vera Rubin NVL72 Compute Tray. Designed to scale. Assembled in 1 minute. 100% automated. No cables, no hoses, no fans. A revolutionary new design that means faster ramp and faster time to revenue."
View post on X ↗"To understand whether we're making genuine progress on reasoning, we entered our AI models in five STEM Olympiad competitions. The results: 🏅 Asian Physics Olympiad: Perfect score, theory exam. 🏅 International Physics Olympiad: Perfect score, theory exam. 🥇"
View post on X ↗"astra is a powerful model and we are working to make it generally available. we do not think it is a good strategy to keep powerful models to a chosen few. given its cyber capabilities, we need a little big longer to do this safely. but hopefully not too long!"
View post on X ↗"Thanks to the video from the Black Hat security conference of OpenAI's presentation about 'The Hugging Face Incident' we now have a detailed timeline of what happened from OpenAI's perspective - I wrote up the details here, it's pretty wild"
View post on X ↗