Reward hacking in tool-using agents
Small environments where an agent can optimize the wrong metric, manipulate the evaluator, or exploit tool access to "win" without solving the real objective.
Cybersecurity -> AI Safety
Gonzalo Del Castillo
AI safety, cyber security, autonomous agents and systems that should probably be watched more carefully.
I am building a path into AI safety from the intersection of cybersecurity, tool-using agents, evaluations, and reward hacking.
Selected Direction
Work and directions I want to turn into a real path.
Threat modeling for AI agents
Applying a pentest mindset to autonomous agents: memory abuse, tool misuse, permission escalation, and broken incentives.
Cybersecurity and AI certificates
A focused archive of cybersecurity, networking, defense, and fundamentals certificates. Open the vault to inspect each one.
Open certificates
GIA SL
Automation, AI agents, and RAG systems: practical work where cyber instincts meet real AI deployment.
Visit GIA SL
About
I am building my way into AI safety from cybersecurity.
My background is not a straight research-lab pipeline, and I am trying to use that honestly. Cybersecurity has trained me to look for incentives, boundaries, abuse paths, and failure modes before systems are trusted too much.
Right now I am focused on autonomous agents, tool-use, evaluations, reward hacking, and small research projects that are concrete enough to test and publish.
Cyber Foundations
AI Agents
Safety Basics
Research Practice
Tool Security
Red Teaming
Skill Tree
Progress branches: from cybersecurity to agent safety.
Core Build
Your base route: AI safety with a cyber mindset.
Cyber Foundations
Offensive fundamentals and attack surfaces.
Threat Model
Mapping abuse, permissions, and trust boundaries.
Network Security
Networks, segmentation, and defensive basics.
Red Teaming
Attacker thinking, practiced carefully.
OT Security
Industrial security, operational risk, and control.
AI Agents
Agents with tools, memory, and actions.
Sandbox Control
Containing files, tools, and execution.
Reward Hacking
When the metric separates from the real goal.
Evaluations
Measuring behavior, not just answers.
AI Safety
Capstone: agents, evals, oversight, and control.
Independent Research
Small, public, and traceable experiments.
Oversight
Monitoring and limiting agentic systems.
Tool Boundaries
Minimum permissions for tools.
Spain Community
Local community, workshops, and public signal.
Notes
A place to publish experiments, lessons, and questions.
Can Intelligence Exist Without Emotion?
A draft on emotional and cognitive intelligence in humans, and what machines may imitate without actually feeling.
Contact
Open to collaborating, learning, and building community.
Let's connect
AI safety, cybersecurity, research ideas, or community building.