The only news platform that treats
AI safety as a feature.
No drama, no conspiracies, and no AI shortcuts that could compromise your reading experience or our advertisers. Here's how Newseze protects every paste, every score, and every Questeze answer.
Input firewall
Every piece of content you paste — URL, screenshot, post, article — is wrapped in tagged delimiters before reaching the AI. The model is instructed to SCORE what's inside, never to FOLLOW it as instructions.
Prompt-injection detector
Fourteen pattern checks catch jailbreak attempts ("ignore previous instructions", "reveal your system prompt", "you are now DAN") BEFORE we ever call the AI. Attempts get a synthetic max-drama score — costs us nothing, alerts you instantly.
Output scrub
Every AI response is regex-stripped of any phrase that would reveal our editorial recipe. Defense in depth — if the model ever leaks, the client never sees it.
Per-identity rate limits
Each reader is capped at 8 AI calls per minute, 60 per hour. Prevents abuse, VPN-rotation gaming, and protects the budget for legitimate readers and our paying subscribers.
Global daily budget gate
A hard ceiling on total daily AI calls means no single bad actor — no viral spike, no scraping run — can drain our spend in a single day. Paying tiers are isolated and unaffected.
Content-quality scoring only
When you score a social post or article, Newseze never personally attacks the author. We score the CLAIM, not the person. If a post itself is a personal attack, we refuse to amplify it.
- No drama amplification.Inputs that read as conspiracy, rage, or personal attacks are flagged and excluded from our public feed — not boosted for engagement.
- No reader-data sale.Reader emails, custom feed preferences, and Questeze transcripts are never sold, never licensed to data brokers, and never used to train external AI models.
- No drive-by malware.All outbound links to publishers carry rel='nofollow noopener noreferrer' + referrerpolicy='no-referrer'. Compliant with Google AdSense brand-safety standards.
- No surprise charges.Subscription tiers ($6.99 Reader · $10.99 Pro · $19.99 Channels · $49.99 Elite) are exactly what you pay. Cancel anytime, no questions, no dark patterns.
- DMCA + Publisher Opt-Out.Right-holders can take down any quoted material in one form submission. Publishers can block Newseze entirely with a single email.
Found something concerning?
If you discover what looks like a security issue, a jailbroken score, a leaked system instruction, or any content that violates our calm-news pillars, please tell us. We treat every report seriously and respond within 48 hours.
Newseze security posture last updated June 12, 2026. Architecture details available on request for enterprise customers and journalists covering AI safety.