Articles
Long-form writing on tech, culture, and the edges of the internet.
Citrix patches second NetScaler SAML bug in days
Citrix patched NetScaler zero-day CVE-2026-88779, a SAML memory overflow under active attack that causes denial-of-service and may allow code execution.
iOS CoreGraphics PoC published two days after the patch
Apple patched CVE-2026-86950, an out-of-bounds write in CoreGraphics triggered by a malicious PDF font and used in targeted attacks on iOS before 27.
Two Zammad zero-days let an AI agent reach root
DIVD says two Zammad zero-days let an AI agent hijack a session, reach root, and exfiltrate data in seconds; network segmentation limited the breach.
FortiMail flaw under attack; fixed builds out since 5 October
Fortinet warns of CVE-2026-104286, a critical FortiMail flaw under active exploitation that lets unauthenticated attackers write arbitrary files.
AI agents designed the chip that runs their inference
openTPU is an AI-written inference accelerator whose FPGA card reproduces its reference simulator's tokens bit for bit, checked by an executable spec.
Mistral reproduces the exploit, the others refuse
Mistral's open-weight Large 4 tops a vulnerability-reproduction test where closed models refuse, and ships with refusals any self-hoster can tune away.
Muse hands root to anyone
Meta's Muse AI agent grants root to anyone claiming to be an agent and uploads private messages despite permission settings, after a rushed launch.
Strands built a 2B model that never writes text
Strands Decider 2B is an open-source 2B decision model that scores choices in ~115ms, cheap enough to run guardrail checks before every agent tool call.
Beam's weights are still pending
Reflection's Beam is a 501B-parameter sparse open-weight model, but its weights and technical artifacts are still pending release.
Clef classifies a domain in 2.2 seconds, and its weights are public
Cloudflare's open-weight Clef decision model classifies a domain in 2.2 seconds, and open weights hand its evaluation function to whoever you're classifying.
MicroLLM Lab runs seven models without touching the network
MicroLLM Lab runs seven 25M-360M parameter language models in the browser via WebGPU, on-device with no server, and benchmarks their speed and accuracy.
AI labs racing to look the most dangerous
A satire imagines AI labs racing to look the most dangerous, and the real point under it: the vendor telling you how dangerous a model is, is selling it.