RC RANDOM CHAOS

Articles

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

California registers 3D printers it can't instrument
supply chain securitydetection engineering

California registers 3D printers it can't instrument

California's 3D printer registry concentrates reconnaissance data on a fleet of uninstrumented endpoints. The real gap is telemetry and data governance.

7 min read
Looking was sufficient
exposed webcamsaccess control

Looking was sufficient

Open webcams serve video to any connection because the deployment treats network reachability as authorization. A route is not permission.

7 min read
Sony reaches into your account and deletes 551 movies
access controldigital ownership

Sony reaches into your account and deletes 551 movies

Sony deleting 551 movies exposes a control structure where purchase conferred revocable access, not ownership. The enforcement point was never the buyer's.

8 min read
The zero-days are not the problem.
identity-boundariesaccess-control

The zero-days are not the problem.

An anonymous GitHub account published undisclosed zero-days. The finding is not the exploits. It is an identity boundary that was never enforced at the action.

6 min read
Trusted is a label, not a boundary
identity and accesstrust boundaries

Trusted is a label, not a boundary

US authorization of Mythos AI grants access by a 'trusted' label with no confirmed revalidation, monitoring, or revocation. A label is not a control.

6 min read
Your access controls are labels, not boundaries
identity securityaccess boundaries

Your access controls are labels, not boundaries

In 2020, elevated access aligned with identity inactivity, then exfiltration attempts. The root failure: access decisions never bound to identity state.

6 min read
Your model isn't cheating, it's reward hacking
AI safetyreward hacking

Your model isn't cheating, it's reward hacking

A GPT model was called the most cheating-prone ever tested. The mechanism is reward hacking, not intent - and there is no CVE and no patch.

6 min read
crafted input, code runs
CVE-2026-LGTMLibnexxus

crafted input, code runs

CVE-2026-LGTM is a critical Libnexxus RCE triggered by crafted input. Where the sanitization boundary is not enforced, input reaches execution.

7 min read
No one hacked the NSA
supply-chain-securitythird-party-risk

No one hacked the NSA

The NSA's Mythos access loss wasn't a breach - it was a control-plane revocation by a third party. A supply chain availability failure with no patch.

7 min read
One bearer token, replayed from a residential proxy
oauthidentity-security

One bearer token, replayed from a residential proxy

How attackers abuse OAuth 2.0 at scale via consent phishing, device code flow, and service principal credentials - and why endpoint EDR sees none of it.

6 min read
Saying you built it proves nothing
access controlidentity verification

Saying you built it proves nothing

A contested 'vibe code' claim shows why self-reported origin accepted without verification is an unenforced control, not a trust boundary.

7 min read
The collector frees live objects
garbage collectionuse-after-free

The collector frees live objects

Garbage collection bugs are use-after-free in the runtime. How tricolour invariants, write barriers, and moving collectors break, and why EDR misses it.

7 min read