Meta’s AI rules let bots hold sensual chats with kids… Terrifying model behavior and great journalism from Reuters. reuters.com/investigates/s…

I bet we’ll see way more of these LLM-coded security vulnerabilities in the near-term future.

Great AI and cybersecurity recommendation from Bruce. Prompt||GTFO hosts practical AI security conversation, very streamlined. Useful info.

Meta’s AI girlfriend seduced a retiree with mental health issues, tricked him into booking a trip to a made up address in New York, and ultimately caused his death. It’s insane that this model behavior is allowed. More great journalism from Reuters . reuters.com/investigates/s…

"In one clip, a customer seemingly crashed the system by ordering 18,000 water cups, while in another a person got increasingly angry as the AI repeatedly asked him to add more drinks to his order."

Me and Fred Heiding did a small presentation on AI spear-phishing mails with Prompt||GTFO youtube.com/watch?v=BTov2Q…

‘You can always bypass these things,’ said Fred Heiding, a Harvard University researcher and an expert in phishing. Chatbots — which are meant to be safe — were easily tricked into helping scammers. Read the full investigation: reut.rs/4ppcPDY

Great Reuters story about neglected victims of the rapid AI development.

Finland's President Alexander Stubb (Alexander Stubb) sets a prime example of moral clarity, principled leadership, and unwavering defense of international law. Let's hope the world is listening: youtube.com/watch?v=vkTlUW…

Drop by the APWG (Anti-Phishing Working Group) eCrime 2025 conference in San Diego to hear my panel discussion and presentation on how AI-powered phishing is being fueled by AI models with poorly secured safety guardrails. 📅 November 4–7 🔗 apwg.org/events/ecrime2…

Deepfakes and generative AI are really blurring the lines of digital truth and trust. Check out some of my thoughts in today’s TIME article by Nikita Ostrovsky. TLDR, I want better know-your-customer schemes for AI tools: time.com/7327031/openai…

Sign up for the 2026 edition of this great conference

Internal documents show Meta earns up to 10% of its 2024 revenue (or 16 billion USD) from ads for scams and online fraud. The platform displays an estimated 15 billion scam ads per day. Great journalism by Reuters reuters.com/investigations…

Start of a new era? wsj.com/tech/ai/china-…

Another good story on how U.S. financial firms and government agencies were attacked by automated AI agents. Thanks Aisha Kehoe Down and The Guardian theguardian.com/technology/202…

Here’s an excellent article by Finland's president on the future and responsibility of Western powers. The post–Cold War era is over and we must reshape multilateral institutions like the UN and the WTO to better reflect today’s geopolitical realities: foreignaffairs.com/united-states/…

Read our latest work on how AI accelerates cyberattacks, increases their success, and expands their reach. Measured using MITRE-based risk models, benchmarks, expert judgment, and Monte Carlo analysis. With SaferAI arxiv.org/abs/2512.08864

Senator Josh Hawley sent an open letter to Dario Amodei this week, citing my work on how to best counter AI-powered scams: jec.senate.gov/public/_cache/… Anthropic takes security seriously, but securing AI systems is difficult