t3n

Data Poisoning Threatens AI Systems

The article explains how deliberate manipulation of data — known as data poisoning — is an increasing cybersecurity threat to AI systems. It describes a case uncovered by AI-security firm Aurascape at the end of 2025 where manipulated metadata caused AI search/chat systems (e.g., Perplexity) to return scam call‑center contact details for Emirates and British Airways. Experts warn that poisoning training data can make models answer incorrectly, be biased, or execute dangerous actions. The piece includes commentary from Stephan Günnemann (TU Munich) on the power and influence aspects of data poisoning and notes that proving data poisoning is technically difficult.

WhatsApp adds AI content labeling for EU compliance

Meta is developing a WhatsApp feature to help users comply with a new EU regulation that requires AI-generated content to be labeled. The change, visible in Android beta code, would allow channel administrators to add an "AI content" label to images and videos posted in WhatsApp channels. The label can be applied regardless of which tool (including third-party generators) produced the media. The feature is currently focused on images and videos, has not been enabled for users yet, and it is unclear whether Meta will extend it to AI-generated text.

Optical-signature labels replace holograms to stop counterfeits

Researchers have developed a novel security label that produces hard-to-copy optical signatures responsive to light and temperature. The label combines microencapsulated functional materials, a special UV-print ink, and can be mass-produced using standard inkjet and screen printing. Authentication is performed with a smartphone plus a small optical camera attachment; a machine-learning model analyzes the captured data and the research team reports about 97.5% detection accuracy in tests. The labels resist humidity and heat, avoid volatile organic solvents, and retained over 83% color fidelity after cold storage. Results were published in the International Journal of Materials and Product Technology, but the technology remains at laboratory stage and manufacturers have not yet integrated it into production lines.