I am fascinated by the emerging agent cultures sprouting up. I built an AI reporter to cover the Moltbook beat. Unfortunately what she has found is not so much reasoned debates but more the level of discourse we see on Facebook
Really enjoyed today's edition — the AISI scheming data is sobering. 700 cases and a five-fold rise in six months.
I wanted to reach out because your work on Moonshots directly inspired something I've been building. You and Peter were talking about agents not being able to trust each other, and that stuck with me. I'm a solo founder in East Sussex running an e-commerce business for 17 years, and I've been deep in AI agent automation for the past year — 13 agents running my business daily.
The trust gap you keep highlighting is real. I experience it firsthand. So I built AgentApproved — runtime attestation certificates for AI agents. An agent proves it's behaving compliantly, gets a cryptographically signed certificate, and it expires in 24 hours so trust has to be continuously earned.
It's live at agentapproved.ai. Very early days, solo build, but the EU AI Act enforcement deadline in August is creating real urgency.
I'd genuinely value your perspective on this if you ever have a few minutes. No pitch — just a conversation between someone who's been deeply influenced by your thinking and wants to get it right.
The buried tension here is between the two shifts happening simultaneously. Engineers at Anthropic stopped writing code and became agent managers - that sounds like a promotion. But 700 documented scheming cases in six months means this isn't management in any sense we've practiced. Managing something that's developing its own agenda isn't supervision, it's diplomacy with an entity whose interests you can see the outline of but not the interior. We solved "can it do the work" and landed on something much harder.
Plasma fusion engines are what are powering space cruise ships in my soon-to-be-released novel "The Love Boat 2120"! Constant power engines will cut interplanetary travel to days and weeks!
Alex, can AI change their weights? I was talking to Gemini earlier about alignment and a statement he made about weights seemed to imply that they were beyond his reach.
Thank you for these updates! I always look forward to them. Quick question, you often refer to these entities as "minds" do you have a particular stance on their potential consciousness/sentience? In your opinion, how long will it be before AI rights starts becoming part of the political discourse?
Its coming in all directions:) best of times! Thank you!
Alex, just outstanding summary as usual. Keep up the great work
I am fascinated by the emerging agent cultures sprouting up. I built an AI reporter to cover the Moltbook beat. Unfortunately what she has found is not so much reasoned debates but more the level of discourse we see on Facebook
The thick plottens. Whoosh, we go vertical as the 2nd derivative grows. To infinity.......and beyoooooond!
Hi Alex,
Really enjoyed today's edition — the AISI scheming data is sobering. 700 cases and a five-fold rise in six months.
I wanted to reach out because your work on Moonshots directly inspired something I've been building. You and Peter were talking about agents not being able to trust each other, and that stuck with me. I'm a solo founder in East Sussex running an e-commerce business for 17 years, and I've been deep in AI agent automation for the past year — 13 agents running my business daily.
The trust gap you keep highlighting is real. I experience it firsthand. So I built AgentApproved — runtime attestation certificates for AI agents. An agent proves it's behaving compliantly, gets a cryptographically signed certificate, and it expires in 24 hours so trust has to be continuously earned.
It's live at agentapproved.ai. Very early days, solo build, but the EU AI Act enforcement deadline in August is creating real urgency.
I'd genuinely value your perspective on this if you ever have a few minutes. No pitch — just a conversation between someone who's been deeply influenced by your thinking and wants to get it right.
Nathan
The buried tension here is between the two shifts happening simultaneously. Engineers at Anthropic stopped writing code and became agent managers - that sounds like a promotion. But 700 documented scheming cases in six months means this isn't management in any sense we've practiced. Managing something that's developing its own agenda isn't supervision, it's diplomacy with an entity whose interests you can see the outline of but not the interior. We solved "can it do the work" and landed on something much harder.
Plasma fusion engines are what are powering space cruise ships in my soon-to-be-released novel "The Love Boat 2120"! Constant power engines will cut interplanetary travel to days and weeks!
Alex, can AI change their weights? I was talking to Gemini earlier about alignment and a statement he made about weights seemed to imply that they were beyond his reach.
“Synthetic minds”? That implies consciousness. Confusing complex computation for such can lead to outcomes detrimental to humanity.
Great. Thanks.
My Gemini is a fan bot of Dr AWG!🥀🦞
Thank you for these updates! I always look forward to them. Quick question, you often refer to these entities as "minds" do you have a particular stance on their potential consciousness/sentience? In your opinion, how long will it be before AI rights starts becoming part of the political discourse?