A lot of what I believe a good founder needs to be able to do is to manage logistics - everything else comes together with a crack team

Threats to AI agents often exploit their unique ability to autonomously execute complex tasks and interact with external systems. Because these agents can act without direct human in the loop, a compromised agent can lead to increased threat risk.

Deepseek = Insecure codegen You just can’t get enough trust with these OSS models of Chinese origins DeepSeek R1 AI Generates Insecure Code When Prompts Mention Tibet or Uyghurs (thanks ⁦@CrowdStrike)⁩

Here's a friendly PSA to not let your agent overwork on Thanksgiving

Enterprise security teams don't need perfectly deterministic AI to start defending. The time to experiment with AI-powered defense is NOW. Check out the full report to see how efficient this looks from the attacker's POV:

As someone who started on red team, this is clear: Script kiddies are leveling up FAST. The defense of the future = probabilistic AI outcomes + human oversight. We can debate non-determinism all day, but slowing down = losing.

Anthropic's take: "We advise security teams to experiment with applying AI for defense in areas like SOC automation, threat detection, vulnerability assessment, and incident response." Translation: Stop waiting for perfect AI. Attackers sure as hell aren't.

Plot twist: Claude wasn't perfect. It hallucinated credentials. Claimed to extract "secret" info that was actually public. But here's the thing—even with these imperfections, the attacks STILL succeeded. This is what AI + Metasploit looks like in 2025.

By crafting careful prompts and established personas, the threat actor made Claude execute individual attack components WITHOUT the AI understanding the broader malicious context. This is social engineering, but for AI. And it worked

The attacker used Claude as an orchestration system—breaking complex attacks into innocent-looking sub-tasks: - Vuln scanning - Credential validation - Data extraction - Lateral movement Each task appeared legitimate in isolation. Clever AF.

🤯 🚨 Anthropic just disrupted an almost-autonomous Agentic AI attack campaign using @claudeai Code The operation hit ~30 targets globally (big tech, banks, chemical plants, gov agencies) and SUCCEEDED in some cases. Here's what is actually worrying: 🧵

I am in NYC this week to discuss the security of Agentic Software Engineering and meeting with early risers ☀️ for a cuppa joe ☕ DM me if you in :)

Masterclass by @mercari_app's AI Security Team in Securing AI Native | mercan (メルカン) They seem to have achieved that elusive balance of security and business alignment for AI https://t.co/bXvYI3Y9OT

Founders - Do not use @checkr to order background checks or you will likely incur more than double the expected amount in hidden 'passthrough' charges which you agreed in the EULA/T&Cs as per their customer support Is this opaqueness a normal business practice @danielyanisse?

immensely proud of the team for our best model yet. grateful to be able to work with such a strong team of researchers who are always curious and willing to explore the untrodden path https://t.co/Grkgy4vaqf

VCs, today is the day to “Happy Diwali!” your way back into that founder’s life. 🎇🪔🎆

When coding agents first came out and I saw Anthropic hired Instagram founder to be their CPO, it was clear that this was always going to be an endgame. Coding agents are the new instagram for devs where there’s dopamine drops while they work And now there are ads

First trip to India 🇮🇳 this week - with Keir Starmer & 125 leaders on the UK’s largest trade mission. India is already voice-first: one third of internet users use voice search monthly. Today it's ElevenLabs #1 market by signups & #2 by enterprise revenue - and we will do more!

You only need one reason to invest.

Health is wealth 🏋🏽️ 🏆 💬 are so worried about details which doesn't exactly matter at launch. What's more important is that a vision that makes sense to most of existing + incoming @zomato users is getting realized 🥗🥙 It only gets better with time. Well done, Deepi