Microsoft Stalker AI Leaks Despite Encryption

Post LinkedIn

🗾Read original on ITmedia AI+ (日本)

#privacy-flaw #security-research #ai-vulnerabilitymicrosoft-stalker-aimicrosoft

💡MS AI security flaw in processes—audit your Microsoft tools now!

⚡ 30-Second TL;DR

What Changed

Microsoft's 'Stalker AI' feature scrutinized

Why It Matters

Encryption protects data itself, but interaction processes have vulnerabilities.

What To Do Next

Test Microsoft AI interaction logs for privacy leaks using tools like Wireshark.

Who should care:Enterprise & Security Teams

Key Points

•Microsoft's 'Stalker AI' feature scrutinized
•Data encryption strong, process vulnerable
•Researchers flag interaction weaknesses

🧠 Deep Insight

AI-generated analysis for this event.

🔑 Enhanced Key Takeaways

•The 'Stalker AI' moniker refers to Microsoft's 'Recall' feature, which captures periodic screenshots of user activity to create a searchable index of past interactions.
•Security researchers identified that the local database storing these screenshots, while encrypted at rest, is accessible to any user with administrative privileges on the device, bypassing standard user-level protections.
•The vulnerability stems from the lack of additional access control layers on the SQLite database file, allowing malicious actors or malware to extract sensitive visual history without needing to break the encryption itself.

📊 Competitor Analysis▸ Show

Feature	Microsoft Recall	Apple Intelligence (Visual Search)	Google Gemini (Contextual Awareness)
Data Storage	Local-first (SQLite)	On-device/Private Cloud	Cloud-based (mostly)
Primary Goal	Persistent activity history	Contextual assistance	Real-time task execution
Privacy Model	User-level encryption	Secure Enclave / Private Cloud Compute	Server-side processing

🛠️ Technical Deep Dive

Architecture: Utilizes a local SQLite database to store metadata and OCR-processed text from periodic screen captures.
Encryption: Employs AES-256 for data-at-rest, but relies on the Windows user's login session for decryption keys.
Vulnerability Vector: The database file lacks granular access control lists (ACLs), meaning any process running with SYSTEM or Administrator privileges can read the decrypted content once the user is logged in.
Processing: Leverages on-device NPU (Neural Processing Unit) for local OCR and image analysis to minimize latency and data egress.

🔮 Future ImplicationsAI analysis grounded in cited sources

Microsoft will implement mandatory hardware-backed encryption for the Recall database.

Public backlash regarding the accessibility of the database by administrative accounts necessitates a shift to TPM-bound keys to prevent unauthorized extraction.

Regulatory bodies will launch formal investigations into the privacy implications of persistent screen-capture features.

The potential for 'Stalker AI' to capture sensitive personal or financial information without explicit per-action consent triggers significant GDPR and CCPA compliance concerns.