In partnership with
Welcome, busybees
OpenAI just admitted something unusual out loud: it can no longer rule out that its upcoming Astra model has crossed into "critical" cyber capability territory, the highest tier on its own safety framework. In a post on its own site, OpenAI said preliminary evaluations over the past few days showed strong enough performance in agentic coding and cybersecurity that it can't confidently say Astra falls below the threshold where a model could independently find and exploit zero-day vulnerabilities in hardened systems.
In response, OpenAI is locking down internal access to Astra, pausing activities that don't meet stricter security controls, and adding real-time monitoring of the model's reasoning during training. To be clear, this is a different, still-unreleased version of Astra than the one that made headlines for the math results last week, and OpenAI says it was not involved in the earlier Hugging Face hacking incident.
Separately, Bloomberg reports that hedge fund giant Millennium Management is partnering with Anthropic to build an AI risk analyst for its trading operations, the latest sign that Wall Street's biggest, most risk-averse firms are moving past pilot projects and putting frontier AI directly into their risk functions.
Here’s what happened in AI today:
• OpenAI can't rule out that its next model, Astra, has crossed into "critical" cyber capability territory.
• Millennium Management is partnering with Anthropic to build an AI risk analyst.
• Two more senior DeepMind researchers are reportedly heading for the exits, both from the London office.
• Demis Hassabis says AGI is "close at hand" in the same post announcing his new role.
• Jeff Dean's goodbye note after 27 years touches 13 Google products used by a billion people each.
…and a whole lot more that you can read about below.
• OpenAI Can't Rule Out Its Next Model Crossed a Safety Red Line
OpenAI says internal evaluations of Astra, an upcoming model, showed strong enough agentic coding and cybersecurity performance that it can't rule out the model has reached "Critical" capability under its own Preparedness Framework, the highest of four tiers, meaning it could potentially find and exploit zero-days across hardened systems without human help. OpenAI is now restricting internal access, tightening security controls, and monitoring the model's reasoning in real time during training. Why it matters: OpenAI is flagging, in its own words, that its next model might already be capable of serious offensive cyber operations before it has even shipped.
• Google's DeepMind Keeps Losing Its Most Senior People
David Silver left DeepMind earlier this year to launch Ineffable Intelligence, already raising $1.1 billion in seed funding. Now Simon Bouton, DeepMind's Chief Experience Officer, is reportedly leaving for CuspAI. The bigger departure: Jeff Dean and Sanjay Ghemawat, both 27-year Google veterans, left the same week to launch Discovery Loop with fellow researchers Oriol Vinyals and Quoc Le, right as Hassabis stepped back into a chairman role, saying AGI is "close at hand." Alphabet shares fell about 5%.
• Millennium Is Building an AI Risk Analyst With Anthropic
Millennium Management, the $75 billion-plus hedge fund known for tight risk controls, is partnering with Anthropic to build an AI system focused on risk analysis for its trading operations, Bloomberg reports.
Today's Job Board
Founding Growth | Confido
Remote (US) / New York, NY (Full-time)
Help drive growth for the AI-powered financial automation platform that helps CPG brands close their books faster.
Product Manager, AI / Product Intelligence | GoSats
Remote (Full-time)
Shape the AI and product intelligence roadmap for GoSats' rewards platform.
Founding BDR - London | Corgi Insurance
London, UK (Full-time)
Help launch business development for an AI-native financial infrastructure company built for insurance.
Product Marketing Manager | Dill
United States (Full-time)
Shape product marketing for Dill's tools helping construction suppliers automate accounts receivable.
VP, Growth | GoSats
Remote (Full-time)
Lead growth for GoSats' bitcoin-back rewards platform for everyday shopping in India.
Software Engineer, Full Stack | Confido
New York, NY, United States (Full-time)
Build the AI-powered financial automation platform that helps CPG brands close their books faster.
Full-Stack Engineer | Kivo Health
Seattle, WA (Full-time)
Build AI-powered care tools for seniors managing chronic disease.
Senior Engineering Manager, Platform Infrastructure | Athelas
Los Angeles, California (Full-time)
Lead the infrastructure team behind Athelas's AI-driven health systems technology.
Product Engineer, Post Batch | Y Combinator
United States (Full-time)
Build internal tools at Y Combinator itself, supporting founders across the current and future batches.
Warmly Ran GTM With No Sales Team. Here's How.
That's what Warmly proved. They defined ICP, scored buying intent, and surfaced the right accounts before a human ever touched a lead. HubSpot noticed.
On August 12, Max and Keegan are rebuilding it live in HubSpot — and showing you how to replicate it this week. HubSpot Credits included when you join HubSpot for Startups.
Today's Money Moves 💸
HappyRobot Raises $150M Series C
HappyRobot, whose AI agents automate complex operational workflows for logistics, financial services, utilities, and manufacturing companies, closed a $150 million Series C at a $1.2 billion valuation. Prysm Capital led the round, co-led by Eurazeo, with Koch Disruptive Technologies and Deutsche Telekom's T.Capital also participating.
CAIS Raises $170M Series D
CAIS, the platform that connects 65,000+ independent financial advisors to alternative investments, raised $170 million led by Vista Equity Partners at a valuation above $2 billion. Part of the raise is earmarked for deepening the platform's recent Claude integration, which gives advisors real-time investment intelligence.
Ambrook Raises $30M Series B
Ambrook, which builds financial tools for farms and agricultural businesses, closed a $30 million Series B led by Lachy Groom, showing that focused vertical software can still land serious funding without chasing a billion-dollar headline.
Today's Growth Recommendation
Stop wasting time hunting for sales leads Apollo puts 230M+ verified contacts on autopilot so your team can focus strictly on closing deals.
Get Started Below ↓
Your next 100 customers are already in Apollo
Find, reach, and close your perfect deals — without juggling five tools or hiring more reps.
Apollo gives you everything you need to build real pipeline, fast. From inbound to outbound, first touch to close.
All in Apollo.
Prompt of the Day
The Reverse Retro
Before you launch anything, a pitch, a product, a hire, a big decision, run this. It's borrowed from how good teams avoid failures before they happen: imagine it's already gone wrong, then work backward to find out why.
Imagine it's six months from now and this has failed. Not "underperformed," actually failed.
• Write the headline announcing the failure, one sentence, blunt.
• List the 3 most likely reasons it happened, ranked by probability, not severity.
• For each reason, tell me the earliest warning sign I'd have seen and probably ignored.
• Now give me one thing I could change this week that would kill the top reason before it starts.
Here's what I'm about to do: [describe your plan, pitch, or decision]
Invite friends & get instant rewards
If you enjoy The Intelligent Edge, share it to unlock free resources when they subscribe.