[ 02 ]
ARTICLES
Writing on model integrity.
What actually changes when you fine-tune, quantize, or ship a language model — and how to know before your users find out.
PINNED
[ 02 ] · June 2026
READ →

Model Integrity Testing Is Not Red Teaming, Evals, or Guardrails
Three reasonable guesses. All three wrong in instructive ways.
READ →
[ 18 ]
September 2026
READ →

Your Model's Name Isn't a Security Guarantee
Hugging Face namespaces can be re-registered by anyone once the original owner deletes their account. Researchers proved it with a live reverse shell.
READ →
[ 17 ]
September 2026
READ →

AI Safety Evaluation vs. Adversarial Robustness: What OpenAI's Astra Gets Right (and Where It Gets Hard)
Researchers are still uneasy about it. Here's the gap between passing a test and being trustworthy.
READ →
[ 16 ]
August 2026
READ →

The Harness Assumes the Model Holds Still
A harness limits what a failure can reach. It does not tell you the failure is new.
READ →
[ 15 ]
August 2026
READ →

Your Red Team Tested a Model. Your Hospital Deployed a Different One.
A peer-reviewed review gives the problem a name: temporal mismatch risk.
READ →
[ 14 ]
August 2026
READ →

AI Is Scaling Where It Is Easiest to Govern, Which Is Not the Same as Where It Is Safest
The governance budget already exists. It is being spent on watching, not testing.
READ →
[ 13 ]
August 2026
READ →

Your Buyer Is About to Ask Who Tested the Model You Shipped
Governance is turning into a due diligence step, and most of the standard questions have no good answer if your model was modified after release.
READ →
[ 12 ]
August 2026
READ →

The Refusal Test Everyone Is Running Is a String Match on the First 128 Tokens
Nobody is validating these models, including the people releasing them.
READ →
[ 11 ]
August 2026
READ →

Capability Evaluations Are Not Safety Evaluations
Your model still works. That is not a safety result.
READ →
[ 10 ]
August 2026
READ →

Quantization and Safety Drift: Why the Model You Ship Is Not the Model You Tested
The artifact gap is one of the least examined steps in the deployment pipeline.
READ →
[ 09 ]
August 6, 2026
READ →

AI Model Hacked During Testing: Why the Harness Is the Real Risk
As models become more agentic, eval environments and control layers are becoming the primary failure point.
READ →
[ 08 ]
August 5, 2026
READ →

Red Hat's asago and the Open Question of Lifecycle Testing
Red Hat's new open source AI governance project automates policy-to-deployment. A look at what it covers, and at what the research literature actually says about safety behavior after fine-tuning and quantization.
READ →
[ 07 ]
August 2026
READ →

Your Fine-Tuned Model Is Less Safe Than the One You Started With
You just haven't measured it yet.
READ →
[ 06 ]
July 2026
READ →
Post-Training Isn't Done When the Loss Curve Flattens
A convergent loss curve means training stopped — not that the model is ready to ship.
READ →
[ 05 ]
July 2026
READ →

From Base to Fine-Tuned to Quantized: A Lifecycle View of Model Integrity
Integrity isn't a property of a checkpoint. It's a property of a lifecycle.
READ →
[ 04 ]
July 2026
READ →

Quantization-Aware Safety Drift: Why INT4 Breaks Your Aligned Model
Aligned models can lose their refusals at INT4 — while every capability benchmark stays flat.
READ →
[ 03 ]
June 2026
READ →

The Release Gate Your Model Pipeline Is Missing
Code does not reach production without passing tests. Models do.
READ →
[ 01 ]
June 2026
READ →
Why Base-Model Benchmarks Fail After Fine-Tuning
The benchmark describes a checkpoint that no longer exists.
READ →
START FREE
If any of this describes your pipeline, SichGate runs the adversarial battery and gives you the differential before you ship.
START FREE ASSESSMENT →