Skip to main content

Threat Model

Proof of Human Intent does not try to detect every threat on the device. It binds a specific action to a human validation and produces a cryptographic proof that malware and AI cannot fake — even when the session, screen, or agent has already been compromised.

This multi-layered security is effective against sophisticated threats, such as:

Scams and Social Engineering​

An estimated $17 billion was lost globally to crypto scams and fraud in 2025, with the average scam payment growing 253% year-over-year to about $2,764. Impersonation tactics and AI-generated scams are overtaking cyberattacks as the primary method criminals use to steal funds, with a reported 1,400% increase in impersonation scams and AI-enabled scams proving 4.5 times more profitable than traditional ones.

  • No credentials to steal or reuse
  • Screen always shows the real action to be executed
  • Human-only readable confirmations

Prompt Injection​

Hidden instructions in websites, emails, and documents can hijack AI agents into sending payments. Prompt injection is ranked #1 on the OWASP Top 10 for LLM Applications, with direct attacks succeeding more than 79% of the time in 2026 testing — and live campaigns have already tricked agents into crypto transfers. Interstellar keeps the human in the loop:

  • Critical actions still require cryptographic human approval
  • The screen shows the real destination and amount, not the injected instruction
  • Visual one-time codes are human-only readable — agents cannot decode or auto-approve them

Clipper Malware​

Warned about by Binance and others, these attacks change transaction details via clipboard manipulation. Interstellar makes this ineffective because:

  • Transaction data is cryptographically secured
  • Approval requires decoding a visual one-time code

Overlay Attacks​

Banking and financial apps are often tricked with fake UI overlays. Interstellar’s dynamic visual cryptography prevents this:

  • An attacker would need to mimic 60–120 fps image changes in sync
  • Computationally infeasible and human-only decipherable