Agent Step-Up: transaction-level cryptographic confirmation of agent side effects

Evidence record

Research prototype binds actor, destination, operation and arguments to a signed approval digest, with controlled attack trials, formal models and measured approval-fatigue limitations.

Read the original source ↗

Claim-level citations (3)

• supportsAgent Step-Up verifies the signed effect instead of a mutable approval narrative: A trusted confirmation mechanism binds a signature to the canonical execution effect and validates the digest immediately before dispatch.

  Protocol design, signed effect object and gateway verification

• supportsAgent Step-Up verifies the signed effect instead of a mutable approval narrative: The study reports forged-narrative success falling from 54% to 3% in the studied setup, but modeled fatigue may still lead to 45% to 60% of malicious actions approved.

  Attack pairs, intervention rates and fatigue limitation

• supportsAgent Step-Up verifies the signed effect instead of a mutable approval narrative: In the paper's ten-runtime audit, none produced a third-party-verifiable approval record; four of eight tested runtimes allowed some form of postapproval effect substitution and none of seven measured runtimes capped approval flooding within a run.

  Abstract; Section 4.2, runtime audit findings, denominators 10, 8 and 7