Security and licences

What never goes into a prompt, secrets, hallucinated packages, insecure patterns AI repeats, scanning, licences of copied code, data settings and your employer's or university's AI policy.

Never paste this

Also called sensitive data, confidential

What: The list of things that never go into a prompt, a file the tool can read or a screenshot you attach, whatever the tool's privacy settings say.
Why: Prompts leave your machine. One pasted production key or customer record can be a breach and a firing offence.

Code

API keys, tokens, passwords, private keys (.pem, id_rsa)
.env files and cloud credentials
Customer or patient data, real emails, phone numbers
Production database dumps or connection strings
Employer code if policy forbids it
Exam, assignment or take-home questions where AI is banned
Anything under NDA

Secrets out of code and context

Also called environment variables, secret scanning, .env

What: Keeping secrets in environment variables or a secrets manager, listing .env in .gitignore, blocking it from the tool and scanning commits for leaks.
Why: AI tools happily hard-code a key they saw to make something work. Scanners catch it before it reaches a public repository.

Shell

echo ".env" >> .gitignore
gitleaks git -v            # scan history for secrets
gitleaks dir .             # scan the working folder
# Leaked? Rotate the key first, then clean history

Watch out

Deleting a leaked key from the latest commit and thinking it is gone: it is still in Git history, so rotate it.

Hallucinated packages

Also called slopsquatting, package hallucination, typosquatting

What: Package names a model invents that do not exist, or that attackers have since registered with malicious code because models keep suggesting the same fake names.
Why: One npm install or pip install of a fake package runs someone else's code on your machine. Every new dependency needs a check before install.

Watch out

Letting an agent in an auto-approve mode install packages it chose itself.

Checking a new dependency

Also called npm audit, pip-audit, supply chain

What: Confirming a package is real, maintained, widely used and needed, and auditing known vulnerabilities, before adding it.
Why: Every dependency is code you trust and ship. Many AI suggestions add a package for something the standard library already does.

Shell

npm view left-pad repository time.modified
npm audit                  # known vulnerabilities
pip index versions requests
pip-audit                  # Python, from PyPA
# Check: official repo, recent releases, downloads,
# and whether the standard library already does it

Insecure patterns AI repeats

Also called SQL injection, XSS, insecure defaults

What: Classic flaws that appear often in generated code: SQL built by string concatenation, unescaped HTML, eval, disabled TLS verification, CORS set to allow everyone, weak password hashing and missing authorisation checks.
Why: Models learned from public code, including insecure code. Knowing the list lets you spot them in review in seconds.
Example: Generated: db.query("SELECT * FROM users WHERE id = " + id). Safe: a parameterised query such as db.query("... WHERE id = $1", [id]).

Watch out

Asking the tool to "act as a security expert" and assuming the output is now secure: tell it the concrete rules instead.

Static analysis on AI code

Also called SAST, linter, Semgrep, CodeQL

What: Automated tools that scan source code for bug and security patterns without running it, run locally and in CI on every pull request.
Why: They check every line at machine speed, including the lines you skimmed. Ask the agent to run them and fix findings as part of done.

Shell

semgrep scan --config auto   # community rules
npx eslint .                 # JavaScript/TypeScript
ruff check .                 # Python
# In CI: run on every pull request

Licences and copied code

Also called open source licence, code referencing, attribution

What: Generated code can occasionally match public code closely, and that code may carry a licence with conditions such as attribution or sharing your changes.
Why: Licence problems surface late and expensively. Some tools can flag or block suggestions that match public code; know your tool's setting and your employer's rule.

Watch out

Pasting a large, distinctive block from an AI answer into a closed-source product without asking where it came from.

Data use and privacy settings

Also called privacy mode, training opt-out, data retention

What: Account and organisation settings that control whether your prompts and code may be kept or used to train models. They differ by tool, plan and whether you use a personal or company account.
Why: The same tool can be fine for a side project and forbidden for client work. Check the settings before you open a work repository.

Code

# Cmd/Ctrl+Shift+J > General > Privacy Mode
# With Privacy Mode on, code is never used for
# training. On by default for teams; admins can
# enforce it.

Employer and university AI policy

Also called acceptable use, academic integrity

What: The written rules on which AI tools you may use, on what code and data, and whether you must disclose it, set by your employer, client or course.
Why: Breaking them can mean a failed course or a lost job, even if the code was good. When unsure, ask before you start.

Watch out

Assuming that because a tool is installed on a work laptop, it is approved for every repository.

Go deeper