Using AI safely at work
Picks checked October 2026.The rule I work by: nothing from your network goes into a free AI tool. That means configs, logs, hostnames, IPs and captures. Your company's approved tool is a different conversation, and it belongs with your security team. These help you have it.
Where I'd go
ChatGPT and large language models: what's the risk? (UK NCSC)
The UK's national cyber security agency, in plain language: don't put sensitive information in queries to public AI tools, and don't ask anything that would hurt if it went public. A government source helps when you're explaining the rule to your manager.
LLM02: Sensitive Information Disclosure, OWASP Top 10 for LLM Applications
What your security team will read. It covers both sides: what users shouldn't enter, and what the company running the tool should publish about keeping and training on your data. Bring it to that conversation.
Model training privacy setting, Anthropic Privacy Center
Where you turn off model training on Claude's Free, Pro and Max plans. It's an example of what to look up for any tool you use: whether your chats train the model, and where the switch is. Turning it off still doesn't make a free tier the place for your configs.
Start here: read this first
My never-paste rule, and the one public config that's safe to practice on.
Left out on purpose
- AI policy templates. Your company's policy has to come from your company, and a template can make a question look settled when it isn't.
- Redaction tools that promise to make any config safe to paste. One missed hostname or public IP is still a leak, so I'd rather not paste at all.