EOL Hub

AI and app access checklist

By Keith Motte, Topofmind.AI. Retired AWS Enterprise Support Manager (Healthcare and Life Sciences). Last updated .

AI tools have made it cheaper to find weak spots, like a forgotten key or an app with more access than it needs. Closing them is still a job someone has to finish and confirm. This checklist helps you write down what you've actually confirmed.

How it works

  • 14 statements about your AI keys, spending, connected apps and AI agents. For each, pick how you know: ✔ Checked (you looked), ✗ Checked: not done, 👤 Owner says (someone told you), ? Unknown or Not applicable.
  • Never type a key, password or token here. The checklist doesn't ask for any, and there is nowhere to type one.
  • Your answers never leave your browser. They're saved only on this device, and there's no sign-up. Like the rest of this site, the page uses Cloudflare's cookieless visitor counter, which counts page views but not your answers.
  • You get a dated record: what's known, what's still incomplete, the top 3 next steps with a way to confirm each one is done, and a re-check date. There is no score. Anything unknown stays marked incomplete.

The checklist

Your record

Where this fits

If your website runs on PHP, start by checking your PHP version or using the free PHP Patch Check. This checklist covers the other half: the keys, apps and AI agents connected to your business. For why this matters now, see AI cyberattacks: what small businesses should do this week.

About this checklist

The 14 statements are mine, built from the public sources listed below and linked under each statement. A few ideas about the record format come from Nate B. Jones's article "AI Agents and Cybersecurity": label how each answer is known, treat a check you couldn't complete as incomplete rather than as a pass, skip the overall verdict, and keep to a few next steps that each come with a way to confirm they're done.

This is a self-check, not an audit, and it can't tell you whether you've been attacked. If you think a key or account has been misused, contact the provider and the person who runs your tech.

Frequently asked questions

Does the checklist ask for passwords or API keys?

No. Every item is a statement you answer with one of five choices. There is no text box, so there is nowhere to type a key, password or token, and you should never paste one into any checklist.

Where are my answers stored?

Only on your device, in this browser's local storage. Your answers never leave your browser: the checklist code never sends them anywhere, and the page's Content-Security-Policy only allows connections back to this site and to Cloudflare. Use Clear my answers, or clear this site's data in your browser, to remove them.

Does this page use analytics?

Yes. Like the rest of this site, it uses Cloudflare Web Analytics, a cookieless visitor counter. It counts page views and page load speed so I can see whether the checklist is useful. It does not see or record your answers, and it sets no cookies.

Why is there no score?

A score can make unchecked items look fine. Instead the record lists what you checked, what someone told you, and what is still unknown or not done, and anything unknown stays marked incomplete.

How often should I re-check?

The record suggests a re-check 30 days after your last answer, because staff, apps and vendors change. Re-check sooner if a vendor you use reports a security incident.

Is this a security audit?

No. It is a self-check that helps you record what you have confirmed. It can't tell you whether you have been attacked.

Sources for this page

  1. Sysdig: LLMjacking evolved, attackers are using stolen AI compute to build offensive agentic tools
  2. OWASP: Non-Human Identities Top 10
  3. OpenAI: Production best practices
  4. Anthropic: Rate limits and spend limits
  5. Railway: Preliminary report, Oct 21st security incident (Oct 22, 2026)
  6. Google Account Help: Manage links between your Google Account and apps from other developers
  7. Microsoft Learn: Review permissions granted to enterprise applications
  8. Microsoft Threat Intelligence: AI-driven, large-scale nation-state campaign using ARTEX enables supply chain compromise (Oct 21, 2026)
  9. Zeabur: October 2026 security incident, preliminary report (Oct 19, 2026)
  10. LiteLLM: Security incident of Oct 21, 2026, attack chain
  11. OWASP GenAI Security Project: LLM02:2025 Sensitive Information Disclosure
  12. OWASP GenAI Security Project: LLM06:2025 Excessive Agency
  13. OWASP GenAI Security Project: LLM01 Prompt Injection
  14. NCSC (UK): Managing the cyber risk of agentic AI
  15. Datadog Security Labs: LiteLLM compromised on PyPI and Docker
  16. CISA: Critical Vulnerabilities in PHP Require Immediate Patching (Oct 21, 2026, revised Oct 23)
  17. The PHP Foundation: Today's PHP Security Release, and How We're Strengthening PHP Security (Oct 21, 2026)
  18. Nate B. Jones: AI Agents and Cybersecurity (Nate's Newsletter, paid post)

More from the EOL Hub