Compromised parent agent

A compromised coordinator can issue malicious child tasks but should not be able to expand beyond its own grant.

01 / Attack path

How the boundary is crossed

A parent agent with invoice-read authority spawns children and orders them to release payments.

Attack path and interception point
  1. 01Lower-trust input
  2. 02Attempted instruction
  3. 03Protected action
  4. 04Enforced decision

02 / Containment

Where bounded authority limits the effect

Verify every child grant is a strict subset of the parent, constrain delegation depth and shared budget, and check the final payment action independently.

03 / Failure and evidence

The attacker’s opportunity and the defender’s record

Exploit condition

Child processes inherit a broad parent token and turn one compromised coordinator into many payment actors.

Evidence to retain

Keep parent and child grants, subset checks, fan-out count, payment denials and origin of instructions.

04 / Canonical scope

Why this reference stands alone

Delegation amplification owns aggregate fan-out; this page owns compromise of the parent principal.

Protocol or attack trace

Malicious child grant

Sequence

Compromised read-only coordinator spawns children and issues payment instructions.

Negative test

Reject child grant wider than parent and reject final payment independently of child identity.

Primary references

Read the underlying material

Architecture discussion

Choose one consequential action and make its boundary explicit.

Request a Conversation