#auditledger

An email system drafts and completes work overnight without the cloud when the agent runs on hardware the organisation owns, reads mail through a zero-egress perimeter, and seals every action in a pos

AI agents finish overnight work by running on operator-owned hardware inside a Sovereign Intelligence Operating System. They read email, meetings and notes offline, draft replies, assign owners and se

Everything arrives as signed artefacts verified against pinned keys, crosses one controlled import path, is exercised in staging before promotion, and the update event itself is sealed to the audit le

In principle, yes. Records of what an AI system did are documents and can be admitted in civil proceedings in England and Wales, with hearsay generally admissible subject to notice and weight rules. T

With the same discipline as any critical system, applied inside the perimeter. Versioned, hashed and signed backups of model weights, document stores, configuration and the audit ledger, held on media

Crypto-agility is designing an AI system to swap cryptographic algorithms without re-architecting it. Name the algorithm in each signed record so it self-describes, verify by that identifier, and a br

Evidence outlives its cryptography when verification never depends on a single living scheme. Post-quantum signatures under FIPS 204 and 205 seal a hash-chained ledger, algorithm identifiers record wh

No, not noticeably. Post-quantum signatures like ML-DSA and SLH-DSA are larger and modestly slower than classical schemes, but an AI audit ledger signs consequential decisions rather than every token,

A sealed AI audit ledger survives a reboot because its trust anchor is bound to a hardware root of trust, a TPM or secure element, and re-derived through measured boot. The genesis entry is sealed at

You prove when an offline AI action happened by running a local time authority whose signed timestamp tokens are sealed into the tamper-evident ledger, anchored by a monotonic counter that cannot run

Cross-model consensus reduces AI hallucination by requiring several independent models to agree before a high-risk action proceeds, since they rarely invent the same falsehood. In a SIOS each vote and

You know an offline AI model has not been swapped because its weights are cryptographically signed, the machine proves which file it loaded through a hardware attestation, and that load is recorded in