OpenAI Is Giving Hackers a Sharper Knife with GPT-5.4-Cyber—But Only the ‘Good’ Ones Will Get It
GPT-5.4-Cyber is a fine-tuned model with fewer safeguards for defensive work. It's the opposite of what Anthropic did with Mythos.
Archive
Category: Cybersecurity & Privacy
Data privacy, security vulnerabilities, surveillance technology, and cybersecurity threats
The Fed and Treasury Summoned Wall Street’s Biggest CEOs to Talk About One Thing—Anthropic’s Mythos
Bessent and Powell held a closed-door meeting with bank chiefs after Anthropic's latest AI model uncovered vulnerabilities that had gone unnoticed for up to 27 years.
Quantum Computers Have a Brutal Amnesia Problem — and It’s Getting Worse
A new Nature Physics study reveals quantum computers forget earlier computations as noise builds up, placing a hard practical limit on circuit depth.
Someone Just Broke All of Google Gemma 4’s Safety Guardrails — and Made It Way More Powerful
dealignai released an abliterated version of Google's Gemma 4 31B that strips every refusal mechanism, achieves 93.7% HarmBench compliance, and only loses 2% on standard knowledge…
Dmail Network Is Shutting Down on May 15 — Its Farewell Letter Is a Brutal Autopsy of Why Decentralized Email Failed
After five years of chasing the dream of truly private, decentralized email, Dmail Network is pulling the plug. Its surprisingly candid goodbye letter explains exactly why…
LinkedIn Has Been Secretly Scanning Your Browser Extensions and Sending Your Data To Israel
An investigation by Fairlinked e.V. reveals LinkedIn injects JavaScript that probes for 6,222 browser extensions, exposing users' religious beliefs, political views, disability status, and hidden job…
White House App Found Tracking Users’ Exact Location Every 4.5 Minutes via Third-Party Server
The Trump administration's official White House App has been discovered collecting precise GPS coordinates every 4.5 minutes and transmitting them to third-party servers, raising significant privacy…
Iran Hacked the FBI Director’s Personal Email. The DOJ Says It’s Real. Kash Patel Hasn’t Said a Word
Iran's Handala group breached Kash Patel's personal Gmail, published hundreds of emails dating to 2010, and the DOJ confirmed the material appears authentic — while OSINT…
OpenAI Just Released Free Tools to Help Developers Keep Teens Safe on AI. Rewrites Its Rules for How ChatGPT Talks to Minors
OpenAI released two open-source safety models, an age prediction system, and updated behavioral rules governing how ChatGPT treats users under 18 — the most comprehensive teen…
Hackers Now Hand Off Stolen Access in 22 Seconds. Google Just Built AI Agents That Fight Back at Machine Speed
Mandiant's M-Trends 2026 report found hackers can pass stolen access between groups in 22 seconds. Google's answer is an AI-powered SOC that never sleeps.