Papers

This paper examines what happens when the same capabilities behind legitimate personalization systems are pointed at business trust, employee workflows, leadership judgment, and institutional perception. The core argument is that AI changes social engineering from a scripted campaign into an adaptive system: one that can read a person, model their context, choose the right pretext, respond in the moment, schedule follow-ups, and scale that attention across many targets at once.

I chose this topic because I build personalization and agentic systems from the inside. The useful version of this technology helps organizations understand people, communicate more clearly, remember context, and act at the right time. But while working with these systems, I kept seeing the same dangerous symmetry: the mechanisms that make personalization valuable for honest businesses can also make deception cheaper, more patient, more believable, and harder to detect.

The paper tries to add a practical perspective to the cybersecurity conversation. It is not only about phishing emails or deepfakes. It is about the decision itself becoming the attack surface: the wire transfer approved, the vendor detail changed, the access granted, the leader doubted, the market narrative believed. My goal is to help cybersecurity leaders, CEOs, and technology teams understand the shape of this threat early, then translate that understanding into actionable education, workflow controls, public-exposure management, verification practices, and new defensive disciplines such as decision hygiene and TrustOps.

Enterprise AI deploys dozens of autonomous agent nodes across workflows, each acting on the same entities with no shared memory and no common governance. We identify five structural challenges arising from this memory governance gap: memory silos across agent workflows; governance fragmentation across teams and tools; unstructured memories unusable by downstream systems; redundant context delivery in autonomous multi-step executions; and silent quality degradation without feedback loops.

We present Governed Memory, a shared memory and governance layer addressing this gap through four mechanisms: a dual memory model combining open-set atomic facts with schema-enforced typed properties; tiered governance routing with progressive context delivery; reflection-bounded retrieval with entity-scoped isolation; and a closed-loop schema lifecycle with AI-assisted authoring and automated per-property refinement.

We validate each mechanism through controlled experiments (N=250, five content types): 99.6% fact recall with complementary dual-modality coverage; 92% governance routing precision; 50% token reduction from progressive delivery; zero cross-entity leakage across 500 adversarial queries; 100% adversarial governance compliance; and output quality saturation at approximately seven governed memories per entity. On the LoCoMo benchmark, the architecture achieves 74.8% overall accuracy, confirming that governance and schema enforcement impose no retrieval quality penalty. The system is in production at Personize.ai.