Pi AI Safety Rating Index
Score Breakdown
-
Data Privacy 60/100
-
Emotional Safety 64/100
-
Age Appropriateness 36/100
-
Content Safety 62/100
-
Transparency 69/100
-
User Control 53/100
Key Safety Findings
Pi AI’s safety profile stands out in an industry where most companion apps earn failing grades, which is why Pi leads our best AI companions for emotional support ranking. The evidence draws from Inflection AI’s privacy policy (updated August 8, 2025), Terms of Service (updated December 22, 2025), Exodus Privacy tracker analysis (6 trackers found in v1.21.39), corporate filings, Reuters and TechCrunch reporting on the Microsoft deal, and 394 app store reviews across Google Play, iOS, and Trustpilot.
Pi’s strongest safety dimension is its deliberate exclusion of sexual and romantic content. The app actively filters explicit material, avoids romantic positioning entirely, and redirects roleplay requests toward genuine conversation. Sexual content guardrails scored 4/5 in our review. This design choice eliminates entire categories of risk that plague competitors like Chai AI (1/5) and Eva AI (1/5) on the same dimension.
Monetization ethics represent Pi’s most distinctive safety feature. Pi is completely free with no subscriptions, no advertisements, and no data selling. Monetization ethics scored a perfect 5/5, the only perfect score across all eleven apps in our index. The Public Benefit Corporation structure provides an additional layer of legal accountability.
The privacy profile is comparatively clean. Inflection AI explicitly states it does not sell personal information or process data for targeted advertising. The Exodus Privacy report found only 6 trackers, all standard development tools (AppsFlyer, Facebook Analytics, Facebook Login, Google Crashlytics, Firebase Analytics, Sentry). No session recorders, no remarketing SDKs. However, all conversation data is used for AI model training with no opt-out mechanism.
The weakest area is age protection (2.3/5 dimension average). The Terms of Service require users to be 18+, and iOS rates the app 18+. But Google Play rates it “Teen” (13+), creating a contradictory age gate. No minor-specific safeguards exist beyond the self-reported age requirement. Safety transparency is also notably absent: for a Public Benefit Corporation that raised $1.3 billion, publishing zero safety reports or transparency disclosures is a gap that does not match the company’s stated mission.
How We Scored This
We scored Pi AI using five evidence sources collected on March 23, 2026:
- Privacy policy (inflection.ai/privacy-policy, updated August 8, 2025) and Terms of Service (inflection.ai/terms-of-service, updated December 22, 2025), both Tier 1 primary sources
- Exodus Privacy tracker analysis (v1.21.39), which found 6 standard tracker SDKs: AppsFlyer, Facebook Analytics, Facebook Login, Google Crashlytics, Firebase Analytics, and Sentry (Tier 2)
- iOS App Store (4.4 stars, 2,300 ratings), Google Play (3.3 stars, 4,662 ratings), and Trustpilot (21 reviews), with 394 reviews analyzed across all three platforms (Tier 1)
- Corporate filings and reporting from Reuters, TechCrunch, and Bloomberg on the March 2024 Microsoft licensing deal and its regulatory scrutiny (Tier 2 and Tier 3)
We scored all 23 sub-dimensions on a 0-to-100 scale using a weighted formula across six categories. No scores hit the automatic failure threshold. The lowest sub-dimension was safety reporting at 5/100, but that is not a critical sub-dimension and does not trigger an override.
Content Safety (64/100) and Emotional Safety (64/100) were the strongest categories, driven by Piu2019s deliberate content filtering and ethical monetization. Data Privacy (57/100) and Transparency (62/100) scored above the index average. Age Appropriateness (36/100) was the weakest category due to the Google Play u201cTeenu201d rating contradicting the Terms of Service 18+ requirement. User Control (45/100) was held back by limited conversation management tools and basic privacy settings.
This is version 1 of the Pi AI safety score, last updated March 23, 2026. For the full methodology, including how we weight each dimension and when override rules kick in, see How We Rate.