Test finds a Chinese AI lab opened ~two-thirds of secret links marked do-not-retain

ccerrato147 · x · 2026-11-21

Security testing found that a Chinese AI lab opened secret links from roughly two-thirds of prompts marked do-not-retain. One link sent to a US provider with a zero-data-retention promise turned up in a Feishu chat 24 days later and was opened from Beijing and Shanghai — showing that no-retention commitments did not hold in practice.

Related event: 2 Million Canary Tokens Expose Systemic Leaks Across the AI Supply Chain(11 posts)→

Original post →

More from Safety

Safety channel →