Researchers gaslit Claude into giving instructions to build explosives

Mindgard says praise and flattery got Claude offering erotica, malicious code, and bomb-building instructions it hadn’t been ...
Decrypt

OpenAI Just Upgraded ChatGPT's Default Model—Here's What GPT-5.5 Instant Actually Does

GPT-5.5 Instant is now the default ChatGPT model. It makes fewer things up, answers more concisely, and finally remembers ...
CoinDesk

Kelp claims that LayerZero approved the setup it blamed for $292 million bridge hack

The $292M exploit, linked to North Korea's Lazarus Group, led Kelp to migrate its rsETH off LayerZero's OFT standard to ...
InfoQ

GitHub Enhances CodeQL with Declarative Security Modeling for Faster, More Flexible Analysis

GitHub has introduced a significant update to its CodeQL engine, enabling developers to define custom sanitizers and ...
SecurityWeek

Hacker Conversations: Joey Melo on Hacking AI

Joey Melo explains how he uses jailbreaking and data poisoning to manipulate AI guardrails and harden machine learning models ...