Our safety guides rely on what vendors document, what security researchers publish and what government agencies advise. We don't run hands-on tests or try to break agents ourselves, and we don't claim experience we lack. Every product safety claim cites the vendor's own documentation, and anything we couldn't confirm is labelled as unconfirmed.

What sources do we use?

We use sources in this order of priority:

  1. Vendor documentation, such as Anthropic's Duet safety article, OpenAI's dots safety FAQ, Meta's Muse security post and the Grok Bot FAQ. A product safety claim needs one of these behind it.
  2. Government and standards bodies, including NIST, CISA, the UK NCSC, the FTC, the CFPB, the European Commission and OWASP. Our government guidance overview summarises the main ones.
  3. Security research, such as academic papers and reports from security firms. We say when a result comes from a research setting rather than real-world attacks.
  4. Reputable press coverage, for incidents and context. Where a vendor disputes a report, we give both sides.

Social media and forums are used only to find leads, which we then try to confirm elsewhere.

  1. CollectGather vendor docs, safety pages, research and official guidance.
  2. VerifyConfirm each claim against the primary source where possible.
  3. HedgeLabel anything disputed, secondary-only or unconfirmed.
  4. DateMark prices, availability and settings "as of" the month checked.
  5. ReviseUpdate pages when vendors change their rules.

How do we write about risk?

We try to be precise and non-alarmist. For each risk we explain what it is, how it could affect an everyday user, what vendors do about it, and what you can do. We report figures with their context: a prompt-injection rate from a vendor's internal benchmark is not the same as a real-world rate, and we say so. Our safety features comparison shows this applied across agents.

What don't we publish?

We don't publish instructions for bypassing safeguards, step-by-step exploits or ways around regional restrictions. When money or personal safety is at stake, we direct readers to their card issuer or bank, the police, IC3 or Action Fraud, and vendor support.

No. Our pages on laws and regulation explain them in plain language, but they aren't legal advice. For a specific dispute, contact a qualified adviser or the relevant regulator.

Updates and corrections

Every article shows its last updated date. When a vendor changes an approval rule, a hard limit or a data policy, we revise the affected pages. Read more about this site or our privacy policy.