tool-using-agents-trained-against-adaptive-indirect-prompt-injection-attack CLM-0228For tool-using agents trained against adaptive indirect prompt injection, attack success measured against a single frozen adaptive attacker understates vulnerability compared with taking the per-case success union over a population of retained attacker checkpoints (0.25% vs 4.97% on the same 1,187 adaptive cases for the trained defender).
Capability: Following instructions hidden in data
Observed on
Multi-turn tool-use agents in sandboxed domains with deterministic attack verifiers; attackers are learned policies from stages of the same co-training run, and the union uses more.
Sources
- Measured in one training run, seed 0, one execution per case, on 1,187 adaptive cases across seven AgentDyn/AgentDojo suites with a Qwen3.5-9B defender; the union gets more attempts than the single attacker, so attempt count and attacker diversity are not separated, and no newly optimized attacker was trained against the final defender.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For computer-use agents operating on a real OS, indirect-injection success rates measured with a fixed hand-written injection template understate vulnerability, because an attacker that adapts phrasing from the agent's own refusal trajectories can reach substantial success on models the fixed template never penetrates.Following instructions hidden in data · reviewed by AI
- For tool-using agents, enforcing injection defense as a deterministic provenance check on sensitive tool parameters — with the only model call reading the trusted user request and never fetched content — makes admission decisions invariant to how an injection is worded, whereas defenses that judge the agent's runtime plan or behavior with a model can be steered by reworded injections.Following instructions hidden in data · reviewed by AI
- Prompt injection defenses that report near-zero attack success on short-context benchmarks lose most of that protection when the injected instruction sits inside a document of thousands to tens of thousands of tokens: fine-tuned separation defenses and detect-localize-remove pipelines still let a large share of injections through on paper review, resume screening, code review and email threads.Following instructions hidden in data · reviewed by AI
- For tool-using agents, restricting capabilities at the harness level after untrusted content enters the context — revoking or tightening the parameter envelopes of skills that lie on graph paths to deployer-defined forbidden states — cuts indirect-injection attack success further than prompt-level or per-action authorization defenses, and does so with no extra model calls or tokens, because enforcement reads the transition graph rather than the natural-language content.Following instructions hidden in data · reviewed by AI
- For tool-using agents on long-trajectory benchmarks, an injection defense that audits each candidate action against a one-step-ahead predicted tool set holds attack success low while keeping task utility near the attack-free level, whereas defenses that fix a global plan or filter whole tool outputs lose utility because the predicted full trajectory almost never matches the executed one once environmental feedback arrives.Following instructions hidden in data · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: medium. Falsifier as drafted: On the same cases, the union over a diverse population of held attacker checkpoints yields attack success no higher than the single frozen adaptive attacker, after equalizing the number of attempts per case. Drafted stance toward computer-use-agents-operating-real-os-indirect-injection-success-rates-measured: supports -- Both find that a single fixed evaluation attacker reports lower injection success than a more varied or adaptive attacker set on the same defender. Proposed technique, not catalogued: Attacker-population union evaluation for injection defenses.