Security, Science & Hardware: top AI agent use cases
This page groups security work, science, and hardware: looking for vulnerabilities, reverse engineering, math, CAD, and robots. There are 32 use cases in this category, each a job someone reported doing, with a link to the original post. The ones readers have voted for most are listed first.
32 Security, Science & Hardware use cases, from every agent.
Top Security, Science & Hardware use cases
Run an AI agent on a Nokia 110 feature phone
Author reverse-engineered a 48MB-RAM Nokia 110 to run an AI agent on it and automate a few actions.
How it went It works as a prototype. Over mobile data, unplugged, it can answer questions, check the battery, switch the torch on and off, start a call and set an alarm. It is not permanent.
Custom agent
Reverse-engineer binaries in Ghidra
An MCP server exposing 110 Ghidra tools for AI-assisted reverse engineering.
How it went The README now lists 253 tools, not 110. It claims 93% fewer API calls from batching, and says it was refined over hundreds of documented functions. It gives no benchmark of results.
Custom agent3 accounts
Design and build a robotic arm with no CAD experience
A builder used Claude Code with FreeCAD and Claude Vision to design, print, and assemble a 5-DOF servo arm with vision and voice control.
How it went About 30 hours from design to assembly produced four working servos. The gripper is not built, there is no feedback loop, and the pixel-to-world calibration is only approximate.
Claude Code3 accounts
Drive a 7-axis robot arm with natural language
OpenClaw is taught a skill and rules for the NERO arm, then writes and runs Python control scripts from plain-language movement requests.
How it went The article is a setup walkthrough and a demo video link. It reports no measured results, success rates, or example prompts with outputs.
OpenClaw3 accounts
Hunt for unpatched vulnerabilities in SQLite commits
Google's Big Sleep agent reviewed recent SQLite commits starting from a patched bug and found an exploitable flaw, which was fixed before release.
How it went It found a stack buffer underflow in seriesBestIndex, and SQLite fixed it the same day, before any release. Fuzzing for 150 CPU hours failed to rediscover it. The team calls Big Sleep highly experimental.
Gemini2 accounts
Generate a robot arm CAD model from one prompt
A text-to-cad skill let Codex produce a 7-DOF robot arm with kinematics, a GUI and STEP parts in one demo prompt, though not the gripper.
How it went The arm's URDF, kinematics, GUI and STEP assembly were all generated, except the gripper. Fitzgerald says similar work would have taken weeks across half a dozen tools. Commenters questioned the joint limits and parts that wouldn't assemble.
CodexClaude Code2 accounts
Hunt business-logic flaws and open fix PRs
Parallel Devin agents each read part of a codebase, find business-logic and auth bypass flaws, confirm them in a sandbox, then open remediation PRs (vendor description).
How it went On Cognition's own 50-vulnerability benchmark, Devin Security scored 72% recall at $90.23 per run, against 68% for Claude Security at $131.87. This is a vendor benchmark, not an independent one.
Devin2 accounts
Discover room-temperature magnetic semiconductor candidates with agents
Opus 5.5 agents were used to identify two candidate room-temperature magnetic semiconductors, per the post's headline.
How it went Both are predicted to be semiconductors with spin windows far above room-temperature thermal energy. YBaMnFeO₅'s ordering likely collapses at synthesis temperatures. KV[Cr(CN)₆] remains unmeasured for band gap and spin sorting, and the two methods disagree on how water in the sample affects it.
Custom agent
Questions
- Is this a guide to finding vulnerabilities?
- No. The entries describe work people already published, including security research. They are not instructions. The details stay in the linked source.
- How many Security, Science & Hardware use cases are there?
- There are 32 use cases in this category on the site right now. This page shows the top ones, not the full set.
- Which agents show up most for this kind of work?
- The named agents used most often here are Claude Code, Gemini and Claude. Other entries use a custom-built or unnamed agent.
- How does a use case get onto this page?
- It has to be something a person reported doing. We re-read the source before it is shown. Reader submissions are reviewed by a person first. An approved submission is checked against its source and added in a later update. A rejected one is not added.
- How is this list ordered?
- By votes from people on this site. More votes means higher on the list. Ties go to the use case described in more separate accounts.