Blog

Topic

#security

6 articles

Hand-drawn warehouse of sealed sandbox crates, with one narrow network pipe escaping under a wall

DeepSeek DSec runs 3 million agent sandboxes a day. The useful lesson is where the work survives

Leo Kaka
Three permission boundaries for OpenAI Dots: read-only background research, reviewed requested actions, and tasks reserved for a person.

OpenAI Dots can keep working. What can it do—and what counts against your plan?

PiRouter Team
Hand-drawn illustration: a small wooden workshop inside a round stone-walled yard. Every way out, the footpath from its door, a cart track and a thin pipe running along the ground, converges on one small gatehouse in the wall, where an open ledger lies on the counter under a single yellow lamp.

ZCode packed git history for upload, an OpenAI agent got out via DNS: 4 egress checks you can run

Leo Kaka
Dark data card comparing twelve LLM providers' API-data training terms: six state no training by default (OpenAI, Anthropic, Google paid tier, Together, Fireworks, xAI); DeepSeek and Moonshot permit training under their published terms; Mistral and Cohere describe opt-out controls without establishing every API product's default; Groq and Alibaba Model Studio could not be verified from the pages read. The three groups total 6, 2 and 4; the footer distinguishes permission to train from proof of actual training.

Does OpenAI train on your API data? 12 providers, 3 answers, one 30-day default

PiRouter Team
Hand-drawn ink illustration: one grey machine with two doors cut into the same body. The left door stands open and a few figures walk through it into a lit corridor; the right door is closed, fitted with a brass turnstile and a card reader, and through its window the same corridor is visible. One figure waits at the right door holding a yellow card.

Astra's "Critical" cyber threshold, Mythos 5.1 and Gemini 3.8 Flash Cyber: one model, two safety tiers, and API access is now a spec-sheet line

Linden Kern
Path diagram: model tokens become Qwen3 Coder XML, enter the vLLM tool parser, reach a highlighted eval() call and then the host; below it a version bar showing eval() present from v0.10.0 through v0.10.1 and removed in v0.10.1.1

Is the vLLM tool-call parser safe? CVE-2025-9141, the 29-day eval() window, and four checks a gateway owes you

Leo Kaka