Daon — third US patent for agentic AI action authorisation (August 2026)
Tag: S-2026-08-07-daon-agent-authorization-patent Type: article (FinTech Global trade-press relay of a vendor announcement; primary Daon release and patent text not read) Author(s): jhaxell (FinTech Global); quotes Tom Grissen, CEO, Daon Date of source: 2026-08-07 (article; patent issued by USPTO 2026-07-21) Date ingested: 2026-08-10 Authority weight: medium — the patent issuance (date, title, issuing office) is a verifiable factual event, but everything about mechanics and market framing is relayed vendor positioning; patent number not given, primary release not fetched Raw file: /_raw_sources/S-2026-08-07-daon-agent-authorization-patent.md
What it claims
Daon (self-styled “Digital Identity Trust Company”) has been granted a US patent — issued by the USPTO on 21 July 2026, titled “Methods and Systems for Authorizing Invocation of a Tool by an Autonomous Artificial Intelligence Agent” — that the article frames as addressing “one of the toughest production hurdles facing agentic AI within regulated sectors”: controlling whether a particular AI agent is permitted to carry out a particular action against a guarded tool, service, account, or API. It is the third in a run of Daon filings on securing and governing autonomous agents; the article says the trio together covers (1) the bond between a person and their agent, (2) the reliability of an agent’s conduct, and (3) the sign-off needed for individual actions an agent takes.
Mechanically (as described), an authorisation checkpoint evaluates each agent request against three criteria — the agent’s continuing connection to the person it acts for, the soundness of its execution behaviour, and the circumstances of the requested action — and, if met, issues a “digital permission slip” bounding the agent’s activity by approved action type, defined scope, and time restriction. Granted claims are said to extend to brief authorisation windows, restricted delegation tools, caps on rate and transaction volume, binding actions to execution context, attestation-derived evidence, and within-session replay protection. CEO Tom Grissen frames the need as “identity, policy, containment, and evidence at the moment an agent attempts to act”, positioned against “the false choice between blocking autonomous agents and granting them broad standing authority”.
Notable quotes
- “‘Methods and Systems for Authorizing Invocation of a Tool by an Autonomous Artificial Intelligence Agent’” (patent title, per article)
- “the system issues what is described as a ‘digital permission slip’ … specifying an approved action type, a defined scope, and a time restriction” (article body)
- “It requires identity, policy, containment, and evidence at the moment an agent attempts to act.” — Tom Grissen, CEO
- “beyond the false choice between blocking autonomous agents and granting them broad standing authority” — Tom Grissen, CEO
What’s speculative vs. asserted
- Asserted as fact: patent issued 21 Jul 2026 by USPTO; title; third in a series of Daon agent-governance patents.
- Vendor-framed / positioning: that this addresses “the toughest production hurdle” for agentic AI in regulated sectors; the safety framing of the CEO quote; “regulated sectors” as target market.
- Not claimed anywhere: that the mechanism ships in a generally available product; any customer (FS or otherwise); any regulatory-standard mapping (no EU AI Act, SS1/23, SR 26-2, DORA or ISO 42001 mention). A patent evidences claimed IP, not deployed capability.
Topics this feeds
- Model Risk Management and Agentic AI — a fourth articulation of the runtime action-authorisation control point (after vendor tooling, MAS SAFR, and Santander’s in-house doctrine), this time as patented IP from an identity vendor.
- AI Governance Platforms — identity-market entrant claiming (and now legally enclosing) part of the agent-authorization/enforcement locus.
Open questions raised
- What is the patent number, and how broad are the granted claims in the actual patent text? (USPTO lookup outstanding.)
- Does patent enclosure of action-level agent authorisation mechanics constrain the open implementations already on the wiki’s record (ValidMind Atryum, Red Hat asago, Santander Autoguardrails) or the harness-enforcement cohort (Credo Agent Governor, HiddenLayer)?
- Is there a shipping Daon product embodying these claims, and any regulated-FS deployment?