Elena' s AI Blog
  Browse by topic  · Browse by tag

All posts
AI Core▶
AI (124) Generative AI (26) Large Language Models (38) ChatGPT (15) Agentic AI (11) AI Development (12) AI Safety (5) AI Law (11) Ethics (7) Edge AI (2) Weekly AI Signals (45) Robots (1) AI Apps (31)
Machine Learning▶
Machine Learning (13) Deep Learning (6) TensorFlow (13) CNN (1) Transfer Learning (1) Mixed Precision (1) Preprocessing (5) Classification (1) Regression (2) Recommender Systems (1) Kaggle (1) Data (1) NLP (6) Fine Tuning (2) Research (4)
Developer Tools▶
Git (26) GitHub (7) CLI (6) Conda (2) Setup (3) Automation (5) Operating System (1) Mac (6)
Content and Site▶
Blogging (20) SEO (6)
Python Series
Personal and Academic▶
Life (37) History (2) PhD (2) Study (4)
Other Tags▶
claude-code (1)

Agents, Access, and the Confused Deputy Problem


Elena Daehnhardt ·

Running a local AI is private. Giving it the ability to take actions introduces a different class of risk. This post explains prompt injection, the confused deputy problem, and the practical mitigations that hold up in 2026 — with macOS and M1 specifics where relevant. Read more...

Local AI Agents with Cline, Ollama, and MCP


Elena Daehnhardt ·

A practical guide to Cline, Ollama, and MCP — runnable stdio + FastAPI example, where to discover servers, and a short student-oriented note on security vs public directory listings. Read more...

Claude Haiku scored zero. GPT-5.5 scored 70%. The new benchmark explains why


Elena Daehnhardt ·

Seven stories from 22–29 May: OpenAI moved closer to a possible trillion-dollar IPO, Pope Leo XIV published a major encyclical on AI and human dignity, Anthropic's Project Glasswing found more than 10,000 serious vulnerabilities, Illinois passed a frontier AI audit bill, DeepSWE exposed a wider coding-model gap, Cognition raised $1 billion, and GLM-5.1 opened a 400-token-per-second API on Huawei chips. Read more...

The Digital Butler or Trojan Horse? A Privacy Playbook for Persistent AI Agents


Elena Daehnhardt ·

Persistent AI agents can save hours each week, but they also turn hidden prompt injections into real-world actions unless you design strict controls. This guide shows how to harden agent workflows with policy gates, isolation, scoped permissions, and safe auditing. Read more...

Podcast: How can we build trust and safety around AI?


Elena Daehnhardt ·

Lawyer Cláudia Lima Costa is an expert in Artificial Intelligence and has created an amazing podcast that raises pertinent questions about trust and safety in AI systems. I was fortunate enough to be invited to a relaxed discussion where I shared my views on various topics related to AI, such as AI evolution, AI applications, data sources for training models, copyright, data protection, privacy-preserving techniques, and achieving reliable, explainable, safe, and helpful AI. Read more...

Copyright © Elena Daehnhardt
Privacy and Terms