Safer
Link
Malicious content hidden in uploaded documents is now screened
Documents you attach to your builds — text files, CSVs, code, HTML, and more — are now scanned for hidden instructions designed to manipulate the AI. Your builds stay under your control.
- Protects against hidden 'ignore all instructions' attacks in uploaded files
- Covers text, CSV, code, HTML, and other document types
- AI behavior stays predictable and trustworthy regardless of file contents