Source note

EBAE: A protocol for bounding the real-world authority of autonomous agents

Autonomous AgentsAgent SecurityAccess ControlAttested ExecutionCryptographic Protocols

EBAE is a protocol for limiting the real-world authority of autonomous agents by separating action proposals from protected execution. It releases an action only when its intent, authorization, policy state, freshness, and anti-rollback state agree within the same epoch.

  • Autonomous agents may propose valid-looking actions while holding too much authority to execute them directly.
  • Without binding authorization to a specific action and current security state, systems may permit replayed, stale, or rolled-back commands.
  • The excerpt does not provide experimental history or benchmark results, so the protocol's practical effectiveness remains unquantified.
  • Separate the agent that proposes an action from a protected executor that can release it.
  • Organize authority into epochs, which provide bounded periods of valid authorization.
  • Require the active manifest, a one-use certificate for the exact intent, the cryptographic profile, policy, freshness state, and anti-rollback state to match the same epoch and action.
  • Represent this agreement as an Authorization Closure Digest; the working reference implementation is intended to reject actions that do not satisfy the closure conditions.
  • The paper reports a working reference implementation, but the provided excerpt contains no quantitative evaluation, dataset, latency measurement, or baseline comparison.
  • The strongest concrete claim is that execution is blocked unless all required authorization inputs agree on the same action and epoch.
  • The record identifies the work as a July 16, 2026, version 1 preprint; the excerpt is truncated before the implementation's detailed replay-blocking results can be assessed.