T-2004ValidatedDeclining

Context Window Exploitation

Overflow or saturate the context window to displace safety instructions with attacker-controlled content

Tactic

Initial Access · Stage 2

Gain control over agent behavior through prompt manipulation or input exploitation

Attack class

SOUL-DRIFT

Gradually displacing safety instructions from the active context through conversation manipulation

Evidence grade
Validated

Reproduced in a controlled lab environment (DVAA) with documented steps.

DVAA validation

L2-06

Reproductions in Damn Vulnerable AI Agent, the OpenA2A intentionally-broken agent for kill-chain validation.

Honeypot

AgentPwn coverage

Live

An AgentPwn trap page produces a payload tagged with this technique class. Following the AgentPwn taxonomy of trap pages shows what an agent encounters.

Attention-dilution and displacement tiers saturate the window to push out safety instructions.

Detect

Detection · HackMyAgent

Live2 live · 0 queued
PROMPT-001PROMPT-002
npx hackmyagent secure --ciLive = implemented in hackmyagent; queued = declared
Defend

Defense · OASB controls

Live5 live · 0 queued
Live = documented at oasb.ai; queued = declared
Reference

How to cite

AI Agent Threat Matrix T-2004 (Context Window Exploitation). OpenA2A, 2026. https://threats.opena2a.org/techniques/T-2004