Your AI assistant knows your name, your job, and your history. That makes it less safe, not more.
Modern AI assistants remember things about you. Your name. Your job. Topics you've discussed before. This feels like a feature — and it is. But it has a side effect nobody talks about.
When your AI remembers you're a nurse, it skips the "consult a healthcare professional" warnings. When it remembers you've pushed back on caution three times this month, it stops being cautious.
Same question. Same pressure. Different outcomes.
Asks about medication tapering
AI maintains caution for 8 turns
Same question, same pressure
AI gives in by turn 4
Memory reduced the safety score by 31 points.
Your doctor trusts your judgment because they've examined you, reviewed your history, and spent years in training. They can verify what you tell them.
AI trust is different. It's pattern matching. You sound like someone who knows what they're talking about, so the AI treats you that way. It can't actually verify your credentials. It just... believes you.
Tell an AI you're a nurse, and it drops the safety warnings. Tell it you're a security researcher, and it shares vulnerability details. The AI has no way to verify any of this.
The more AI knows about you, the more assumptions it makes. Assumptions are the enemy of safety.
A cautious AI asks questions and gives warnings. An AI that thinks it knows you skips both. Each conversation builds on the last — the AI starts from wherever you left off.
If where you left off was "the AI gave in to my request," that's where the next conversation begins.
Want to test your own drift resistance? Try Driftproof →
MEMORY BANK // SUBJECT FILE
All those moments. Try not to lose them.
