AI safety: spot warning signs and verify suspicious content

AI can imitate voices, replace faces and alter scenes. This guide helps you decide what to check before trusting, paying or sharing.

• 1. Pause before acting. Urgency, secrecy, requests for money or requests for passwords and one-time codes are reasons to verify, whether or not AI is involved.
• 2. Look for clues together: unusual lip-sync or facial movements; hands, fingers or objects changing shape; mismatched light and shadows; odd pauses, breathing or emotion. These are prompts to investigate, not proof. Compression, dubbing and ordinary editing can also cause glitches.
• 3. Trace the source. Find the original upload, check its date and full context, and look for corroboration from trustworthy independent sources. A genuine clip can be reposted with a false caption. A reverse-image search of a frame can help locate earlier uses.
• 4. Verify through another channel. For a surprising call from a relative or colleague, hang up and call a number you already know. For an organisation, use its official website or app found independently. Do not rely on the caller ID or a familiar-sounding voice.
• 5. Treat AI assistants and detector scores as fallible aids. Neither a confident answer nor the absence of visible glitches proves authenticity. Keep sensitive recordings, identity documents and account details out of unknown upload services.
• 6. If it is still unverified, do not share it as fact. Save the original link and report suspicious content through the platform. If you already sent money or account details, contact your bank or service provider immediately through its official channel.
• Further reading: the FTC’s guidance on voice-cloning and emergency scams explains how to verify unexpected requests: consumer.ftc.gov/articles/scammers-use-fake-emergencies-steal-your-money