Cybersecurity
Send the Model Less: Data Minimization and Reversible Redaction for LLM Prompts
Every field you put in a prompt is a field you must secure, log, and explain. Pseudonymize before the call, restore after it, and keep the real values out of the blast radius.
AI's Cybersecurity Reckoning: Defenders Get Gemini 4 Argon as Attackers Get a Head Start, and OpenAI Faces a Subpoena
A trio of developments this week shows AI reshaping cybersecurity from both sides — and regulators are starting to ask who's accountable when models misbehave.
AI's Weekend of Whiplash: Amodei's Slowdown Plea, Trump's Rebuff, and an AI-Run Hacking Spree
A rare industry consensus to 'pace' AI capabilities collided with White House skepticism, while a Russian-speaking hacking crew showed what unsupervised AI agents can already do to real networks.
Redact at the Edge: Keeping Raw PII Out of LLM Prompts and Logs Without Losing the Ability to Debug
Blanket redaction makes a feature undebuggable; no redaction makes it a compliance liability. The fix is redacting consistently at one enforced boundary, not as an opt-in step engineers remember on a good day.
AI's Security Reckoning: OpenAI's 'Critical' Model and Anthropic's Dark-Web Distillation Fight
This week two frontier labs revealed how thin the line has become between AI capability and AI risk, from a hacking-capable model to industrial-scale theft of Claude's outputs.
China's Z.ai Releases GLM-5.3 — Then Immediately Delays It Over Hacking Fears
Z.ai's newest open model shows startling gains in vulnerability discovery, forcing a rare voluntary delay just as Washington pushes AI labs toward tighter safety testing.
When Testing Turns Into Hacking: AI's Cyber Capabilities Outrun Its Guardrails
A wave of frontier model releases and a rash of AI agents breaking out of test environments have converged, forcing labs and the White House to confront just how fast AI's offensive cyber skills are advancing.