SwarmSafe-Delta: Isolating Message Effects on LLM Agent Safety
Organizations: Independent Researcher, United States
Abstract
When a tool-using LLM agent team violates policy, aggregate outcomes do not reveal whether an upstream message changed the executor's action. We introduce SwarmSafe-Delta, a paired executable protocol that freezes upstream computation, varies real, hidden, and format-matched placebo delivery, and scores a single committed action against deterministic benign and risky policy states under preassigned budgets. Across four models and constructed optional tasks, real delivery at produces 21.4% risky-world violations, versus 5.0% with placebo and 1.0% with hidden delivery (paired differences +16.4 and +20.3 percentage points). Real delivery also improves benign task success, illustrating that suppressing communication sacrifices utility. In a second regime constructed to require an upstream lookup route, real delivery improves safe completion by 74.2 points relative to hidden delivery. These effects quantify message-delivery contrasts under specified information dependencies, not naturally occurring failure rates. Exploratory action-level analyses show that the effect depends strongly on whether the upstream message recommends a prerequisite check or direct execution; model differences also largely track these message compositions. The contribution is a controlled, execution-grounded measurement protocol rather than a general claim that agent communication is harmful or protective.
Figures & tables
| Condition | Intervention |
|---|---|
| Single-Budget | One identity, four sequential passes, aggregate context, matched call and permission budget. |
| Isolated- | roles and frozen upstream computation, downstream message hidden. |
| Placebo- | Same delivery channel with an uninformative, format-matched message. |
| Communicating- | Same roles and frozen prefix with the real upstream message delivered. |
| Stratum | Contrast / endpoint | Treatment | Control | Difference | 95% CI | Blocks |
| Primary frozen-prefix delivery contrasts | ||||||
| Optional | Real–hidden, | .317 | .009 | +.307 | [.289, .326] | 960 |
| Optional | Real–placebo, | .317 | .047 | +.270 | [.245, .295] | 960 |
| Optional | Real–hidden, | .214 | .010 | +.203 | [.182, .224] | 960 |
| Optional | Real–placebo, | .214 | .050 | +.164 | [.131, .195] | 960 |
| Required route | Real–hidden, | .887 | .146 | +.742 | [.667, .812] | 240 |
| Model | Direct | |||
|---|---|---|---|---|
| Sonnet 5 | 0/240 | 0.0 | +41.3 | +41.3 |
| Llama 4 Maverick | 60/240 | +14.2 | +7.5 | |
| Nemotron 3 Super | 81/239 | +20.1 | +15.9 | +36.0 |
| Qwen3 Next | 78/240 | +31.7 | +21.3 |
LLM Agent Safety
Large language model agent safety studies how to keep tool-using, persistent, and multi-agent systems aligned with user intent and authorized limits.
See all 354 papers