Articles
Long-form writing on tech, culture, and the edges of the internet.
Cerebras runs Qwen 27B at 1,500 tokens a second
Qwen 3.8 27B on Cerebras at 1,500 tokens/s adds no new capability - it changes the economics of attack and defense. What the raw speed means for security.
Make the binary say yes
Why a stripped binary is not a security control, how reverse engineering recovers its logic, and where access decisions must actually execute.
The sand was never boxed
A sandbox is a set of filters on named dimensions, not a boundary of execution. It matches references and never revalidates the capability they compose.
Verification became the leak
An unauthorized live feed of every ID verification let attackers bypass MFA for over a year. Why readable verification output stops being proof.
Gemini 3.8 Flash captures your evaluation inputs
Evaluating Gemini 3.8 Flash and Flash Cyber transfers your input to infrastructure you cannot inspect. What failed, the pattern, and what must now be true.
Perplexity cites 215,128 template pages as evidence
Three sites generated 215,128 scripted "best software" pages that AI engines like Perplexity cite as sources - fueling misinformation, SEO abuse, and malware.
Three model calls, finished before you blink
Qwen 3.8 27B at 1500 tokens/s on Cerebras is an orchestration budget, not a quality upgrade - spend it on validation and decomposition, not more agents.
FBI probes sale of 153 million license records
An FBI probe into a service selling 153 million driver's license records shows why static identity data cannot be rotated, recalled, or trusted as proof.
Firefox blocks cross-site tracking by default
Firefox moved tracking protection from optional settings into the engine by default. Why enforcement, not preference, protects your anonymity.
Same badge, new model underneath
A systems-analysis read on Claude Fable 5.1 and Mythos 5.1: why a point release quietly moves the safety boundary and what to re-test before you ship it.
The 150x model tax
Cost, latency, and quality trade against each other in LLM inference. How to route work, use serving-layer levers, and architect systems on the efficient frontier.
A robot dog moves on command inside your network
The Creepy Crawlies robot dog is a networked device that moves on command. Its control channel and actuator define the risk, not its label.