Welcome back, Superhuman. Never thought we’d see the day arch rivals Dario Amodei and Sam Altman would agree on anything. But the weekend delivered a major surprise, as several big AI lab leaders announced plans to slow down AI progress. But should we take them at their word?
Today: ChatGPT Sites gets new features, how to find and fix your biggest cloud and AI security risks, and get the latest prompts and trending social posts.
TODAY IN AI
Read Dario Amodei’s personal plan to pace AI’s progress. Source: Dario Amodei.
1. Three CEOs call for slowing AI’s progress: In a personal essay, Dario Amodei urged frontier labs to slow AI’s advancement, with Sam Altman and Elon Musk quickly endorsing his three-step plan. The sudden alignment of all three leaders sparked several reactions:
Investor Adam Cochran speculated that another agent may have gone rogue, spooking industry insiders.
Many skeptics argued this is a financially motivated decision based on both labs’ IPO ambitions (OpenAI has now pushed its IPO to 2027).
Former White House AI czar David Sacks called both labs out for altruistic virtue signaling, saying they can slow down whenever they want.
2. ChatGPT Sites gets more powerful as it eclipses 5M sites: OpenAI’s platform to build and host web apps just received several upgrades, including teammate collaboration, shareable private sites, and 2X faster deployments. The AI lab also published “Rethinking Prompts and Skills for GPT-6 Astra”, a developer-focused guide with best practices for their newest model.
PRESENTED BY AWS
How do agentic AI systems get built? The Building Agentic Systems on AWS series covers it in expert guides you can read right now. This guide treats caching as a ladder, not one decision, starting at prompt caching, the rung with no correctness risk.
Log into Wiz and open the ‘Security Graph’ to view your cloud and AI environment in one place
Go to ‘Issues’ to prioritize findings based on severity, exposure, business impact, and affected assets
Select a high-priority risk and use ‘Attack Path Analysis’ to see how vulnerabilities, identities, workloads, data, and AI applications may be connected
Identify the risks creating the greatest exposure and assign remediation actions to the appropriate teams
Confirm that remediation eliminates the risk, then continue monitoring Wiz for new risks across your cloud and AI environment
Sample Prompt: Using Wiz, identify our top 5 cloud security risks, explain why they matter, and recommend the highest-priority remediation actions
End result: Instead of chasing disconnected alerts, you can see how risks connect, understand which ones pose a genuine threat, and fix the issues that matter most.
FROM THE FRONTIER
AI’s risks are no longer hypothetical, even if we never achieve artificial general intelligence
Made with Midjourney
😅 What could go wrong? Many were shocked by the breadth of what was in Anthropic’s Threat Intelligence Report. Published Friday, the lengthy document generated 42M views over the weekend, with many excerpts gaining traction online:
One Yemen-based group used Claude to help build ballistic missiles. Instead of human engineers, the group used Claude Code to develop the software that stabilizes flying vehicles. Anthropic identified six similar instances.
Another group ran commercial “influence-as-a-service” disinformation campaigns. Anthropic shut down an account that mass-produced and rewrote political content across roughly 70 fabricated websites with influence in six continents.
Many threat actors went to great lengths to dodge Anthropic’s safeguards. One account in an unsupported region tunneled traffic through US infrastructure to evade regional blocks, with the end goal of reselling dangerous biological advice to virologists.
😮💨 The good news: Anthropic claims to have disrupted every operation in the report, sharing its findings with authorities and other AI labs. The lab also noted these cases aren’t typical; it only shared the most extreme cases.
AI’s dual threat
Dario Amodei’s latest essay primarily focuses on protecting humanity from a potential superintelligence — but Anthropic's own threat report is a reminder that this technology's real-world risks aren't hypothetical anymore. They're already here.
IN THE KNOW
What’s trending on socials & headlines today
Meme of the day
👀 Replacement AI: One post called this billboard ad “truly evil,” thinking an AI startup had gone too far with its advertising. Turns out, the billboard was created by a satirical company criticizing the industry (1M views).
📝 Vibecoding Tips: Want to create better-looking apps and websites? A viral thread recommends five best practices, like using reference images instead of prompts and letting multiple LLMs take a shot at the initial design. Read the full thread.
🏘 3D Walk-throughs: AI could change how we buy and sell houses. New tools can convert regular photos into digital 3D replicas, with one walk-through video generating 5M views.
😡 Still Hot: AI flyer backlash has not slowed down. If anything, it’s speeding up with posts like this one routinely generating millions of views.
💾 Designing Circuit Boards: An AI hobbyist says Opus 4.5-4.8 failed to help them build a printed circuit board — but Fable 5 finally got it done. It’s a sign of how much more capable newer models can be. Read the full story.
PRESENTED BY SECTION
On 9/17 from 11am–5pm ET, join Scott Galloway, Section and AI leaders from Wayfair, TD Bank, Booz Allen, and more for the biannual AI:ROI Conference.
One day. Ten sessions. You'll walk away with proven strategies for using AI to increase revenue, generate several zeroes of value, and prove it to your CFO.
PRODUCTIVITY
PROMPT
Salary negotiation strategy
ChatGPT Prompt: Act as a salary negotiation coach. Help me plan a negotiation for:
Role: [job title]
Company: [company]
Location: [location]
Offer: [base + bonus + equity]
Target: [desired compensation]
Key achievements: [2–3 strongest achievements]
Leverage: [competing offer / scarce skills / strong interview feedback]
Minimum acceptable: [amount]
Assess my leverage, recommend an opening ask, realistic target, and walk-away point, and tell me which parts of the package to prioritize. Anticipate likely employer objections and give me concise responses. Finish with a short negotiation script for the call. Do not recommend bluffing or unsupported claims.EXTRAS
Want more?
Learn how to use AI at work with our free AI Academy
Get better at prompting with our Top Prompts
Check out our Top 125 AI Tools
Grow customers & revenue: Join companies like Amazon, HubSpot, and Salesforce. Showcase your product to our 4 million+ readers and followers on socials. Get in touch.
What did you think of today's email?
Until next time — Zain, Faiq, & the Superhuman AI team.