Zeroize Command

Zeroization must target both volatile memory (RAM, CPU registers, cache) and non-volatile storage (SSDs, TPM NVRAM, EEPROMs). For flash storage, simple file deletion is insufficient due to wear-leveling algorithms; the command must issue a secure erase or crypto erase instruction to the storage controller.

• Volatile: Overwrite buffers, stack variables, and heap allocations
• Non-volatile: Trigger ATA Secure Erase or NVMe Format NVM commands
• Must handle wear-leveling and garbage collection edge cases on SSDs

The zeroize command must be atomic—it either completes fully or the system enters a fail-safe error state. Partial erasure is unacceptable. Once initiated, the operation cannot be interrupted by software interrupts, task schedulers, or even higher-priority threads. The command must also be irreversible; no backup key material should persist.

• Disables all maskable interrupts during execution
• Returns a cryptographically signed attestation of completion
• Often implemented in ROM or write-protected firmware to prevent tampering

Software zeroization relies on the CPU to execute overwrite loops, which is vulnerable to compiler optimizations that may remove 'dead stores.' Developers must use volatile memory barriers and explicit_bzero() or SecureZeroMemory() to prevent optimization. Hardware zeroization uses dedicated logic circuits that physically short memory cells or blow fuses, providing higher assurance.

• Software: memset_s() from C11 Annex K, zeroize crate in Rust
• Hardware: One-time programmable fuses, battery-backed SRAM with tamper switches
• Hybrid: TPM2_Unseal followed by TPM2_EvictControl to remove sealed keys

The dynamic and immediate removal of an agent's access rights to specific tools, APIs, or data sources. This is a non-lethal containment measure that preserves the agent process while restricting its capabilities.

• Granularity: Per-tool, per-API-key, or per-resource
• Protocol: Often implemented via OAuth 2.0 token revocation or API gateway policy updates
• Relationship: Permission revocation is a softer alternative to a zeroize command when full data destruction is not warranted

The immediate network and process isolation of a potentially compromised agent, restricting its communication to only a controlled environment for observation and analysis.

• Implementation: Network micro-segmentation, sidecar proxy blocking, and namespace isolation
• Forensic Value: Preserves volatile memory for post-incident analysis, unlike a zeroize command which deliberately destroys it
• Decision Point: Choose quarantine when you need to study the attack vector; choose zeroize when data confidentiality is paramount

A pre-defined, secure condition that an autonomous system automatically enters upon detecting a critical malfunction, designed to minimize potential damage or danger.

• Examples:
  • Robotic arm returning to a locked, upright position
  • Financial agent canceling all pending transactions
  • Database connection pool dropping to read-only mode
• Zeroize Integration: A zeroize command is often the final step in a fail-safe sequence, executed after the system has reached its safe physical or logical state

A safety mechanism that automatically triggers a kill command or safe state if the human operator becomes incapacitated or fails to provide a periodic confirmation signal.

• Pattern: Heartbeat-based; absence of signal triggers action
• Application: Critical for long-running autonomous agents that may outlive operator attention spans
• Zeroize Variant: A dead man's switch can be configured to issue a zeroize command rather than a simple kill, ensuring cryptographic material is destroyed if the operator is compromised