The Ironclad Clinic: Why I Chose Local AI Over the Cloud
In forensic psychology, data isn’t just code—it’s a life. Here is why I severed the connection to Big Tech.
Dr. Sophie Andrews
Principal Clinical Psychologist & Founder, IN7AI
The Digital Confessional
In my line of work, raw material isn’t code or commodities; it is the human mind in its most vulnerable and exposed states.
As a Principal Clinical Psychologist specialising in forensic assessment and neuropsychology, I sit at the intersection of the legal system and biological reality. I assess fitness to plead, assess recidivism risk, and map cognitive decline in patients with acquired brain injuries.
The efficiency promised by Artificial Intelligence is undeniable. The ability to summarise hundreds of pages of case files or look for patterns in cognitive test data is revolutionary. But when the sales representatives pitched me the industry-standard “Enterprise Cloud” AI solutions, I balked.
They showed me encryption certificates. They spoke of “SOC2 compliance.” They assured me the risk of a breach was “statistically minimal.”
I politely declined. To them, “minimal risk” is an acceptable margin of error in a business model. To me, it is a potential breach of my patient’s trust and confidentiality.
Here is why I insisted on running my AI locally—on my own hardware—completely severed from the cloud.
The “Trust Me” Fallacy
The primary catalyst for my decision wasn’t technical; it was historical. We are living in an era defined by the betrayal of digital trust.
When cloud vendors sat in my office, they spoke with the easy confidence of institutions that have not yet had their reckoning. But I read the papers. I watch the news. I have seen the scandals—from credit agencies losing the financial details of millions, to social media firms quietly monetising psychometric profiles, to the steady drumbeat of ransomware crippling hospitals and NHS trusts already stretched to the limit. These weren’t small startups; they were tech giants with “military-grade” security.
The Forensic Reality: If I upload a forensic evaluation of a high-profile defendant to a cloud server, I am technically handing over the chain of custody to a third party. If that company uses my data to train their next model, or if a security patch fails, that data is compromised. In forensic work, a leak isn’t just a GDPR violation; it is a potential mistrial.
The Neuro-Privacy Imperative
In my neuropsychological practice, the stakes are different but equally high. I deal with the biological blueprint of a person’s cognition.
If a credit card is stolen, you can cancel it. If a password is hacked, you can change it. But you cannot change your neuro-cognitive profile.
Data regarding a patient’s executive dysfunction, early-onset dementia markers, or IQ testing results is sensitive and highly personal. Cloud providers often claim ownership of “meta-data” or “aggregate learnings.” I refused to let my patients’ neural profiles become part of a dataset used to train a commercial algorithm.
The only place that data belongs is in my encrypted, air-gapped files.
The Solution: Local Intelligence
So, I took a different path. I invested in a high-performance laptop with significant GPU power. I installed open-source Large Language Models (LLMs) that run entirely offline.
This is my “Air-Gapped” workflow:
- Disconnect: I physically disable the Wi-Fi on my machine.
- Process: I feed the sensitive case data into my local AI.
- Analyse: The processing happens on my silicon, right in front of me.
- Delete: Once the report is generated, the context is cleared.
No packets leave the room. No API keys verify with a server in California.
Is it as fast as the massive cloud clusters? Not always. Is it slightly more complex to set up? Yes. But the peace of mind is absolute. I can look a lawyer, a judge, or a worried parent in the eye and tell them, truthfully: “This data has never left my possession.”
Conclusion: Ethics Over Convenience
We are often told that to embrace the future, we must surrender our privacy. I disagree. As clinicians, we are the guardians of secrets that people are often afraid to admit even to themselves.
The cloud offers convenience, but the cost is a loss of sovereignty. In a world where Big Tech breaches are a question of when, not if, keeping my AI “on-premise” is not paranoia. It is the modern evolution of doctor-patient confidentiality.
I will use AI to help my patients, but I will not let the machine—or its creators—compromise them.