RC RANDOM CHAOS

Articles

Long-form writing on tech, culture, and the edges of the internet.

tanh now fingerprints your OS
browser fingerprintingchromium

tanh now fingerprints your OS

Chromium 148 dispatches Math.tanh to the platform libm, leaking OS identity through last-bit IEEE 754 divergence. A silent, unpatchable fingerprinting signal.

6 min read
The flat line is the exploit
detection-engineeringdefense-evasion

The flat line is the exploit

A flat SOC dashboard can be defense evasion, not a quiet network. How APTs shape telemetry to land in a pipeline's discard paths, and why the graph is the IOC.

5 min read
Train the AI or Samsung erases your health record
data privacyAI training

Train the AI or Samsung erases your health record

Samsung deletes your health data unless you let it train AI. That is not consent. It is coercion recorded as authorization.

6 min read
XBOW topped HackerOne in 2025 and fixed nothing
vulnerability-researchcve-disclosure

XBOW topped HackerOne in 2025 and fixed nothing

Vulnerability research rewards finding, not fixing. Exploitation tracks disclosure, not the flaw. Why attackers react to reported CVEs, and where telemetry goes blind.

6 min read
Possession is the credential
authenticationsystems drift

Possession is the credential

Verified badges, JWTs and X.509 certificates resolve trust once and keep honoring the reference long after the reality behind it has moved.

7 min read
The kernel keeps its word
OpenBSDuse-after-free

The kernel keeps its word

OpenBSD's use-after-free is not a bypass but the kernel honoring a reference it validated once and never rechecked, the same pattern behind X.509 certificate trust.

9 min read
The US bought a cheaper way to keep losing
drone securityidentity boundary

The US bought a cheaper way to keep losing

Iran's destruction of $1B in Reapers exposes predictable targeting as a static-trust failure. A cheaper drone with the same pattern is engaged at the same rate.

6 min read
A smarter model would have leaked it too.
AI agent securityLLM engineering

A smarter model would have leaked it too.

GitHub's AI agent leaked private repos not from a bug but a design failure. How two-plane architecture, scoped tokens, and deterministic validation stop it.

10 min read
Open source is not decentralized
supply chain securityconcentration risk

Open source is not decentralized

Damn Interesting going paid is an indicator, not a failure: deep technical analysis concentrates value, and concentrated value gets targeted.

7 min read
The browser runs whatever the host returns
systems failure analysistrust delegation

The browser runs whatever the host returns

A browser tab holding 2 GB is not a malfunction; it is a trust model that resolves references and never revalidates the referent behind the name.

8 min read
The copy runs past the allocation, again
vulnerability-managementdetection-engineering

The copy runs past the allocation, again

Recurring dystopian tech vulnerabilities persist because defenders patch the CVE instance and never hunt the underlying mechanism. The inaction is the vuln.

6 min read
The kernel is still C
use-after-freeopenbsd

The kernel is still C

An OpenBSD kernel use-after-free (CWE-416, CVSS 7.8) escalates a local user to root via pool reclaim and cr_uid overwrite. Mechanism, exploit path, and the BSD telemetry gap.

7 min read