MIT AI Risk Repository · Risk Sub-Category · 63.03.02

Steganography

Category: Collusion

Open the MIT Risk Navigator All entries from Hammond2025

Description

"Steganography. In the near future we will likely see LLMs communicating with each other to jointly accomplish tasks. To try to prevent collusion, we could monitor and constrain their communication (e.g., to be in natural language). However, models might secretly learn to communicate by concealing messages within other, non-secret text. Recent work on steganography using ML has demonstrated that this concern is well-founded (Hu et al., 2018; Mathew et al., 2024; Roger & Greenblatt, 2023; Schroeder de Witt et al., 2023b; Yang et al., 2019, see also Case Study 5). Secret communication could also

From Multi-Agent Risks from Advanced AI (Hammond2025), as extracted by the MIT AI Risk Repository (CC BY 4.0).

Classification

Domain: 7. AI system safety, failures, & limitations
Subdomain: 7.6 Multi-agent risks
Causal entity: AI
Intent: Intentional
Timing: Post-deployment

Subdomain definition: Risks from multi-agent interactions, due to incentives (which can lead to conflict or collusion) and/or the structure of multi-agent systems, which can create cascading failures, selection pressures, new security vulnerabilities, and a lack of shared information and trust.

How other frameworks describe this risk

Groups of LLM-Agents May Show Emergent Functionality
Foundational Challenges in Assuring Alignment and Safety of Large Language Models (Anwar2024)
Collusion between LLM-Agents
Foundational Challenges in Assuring Alignment and Safety of Large Language Models (Anwar2024)
Multi-Agent Safety Is Not Assured by Single-Agent Safety
Foundational Challenges in Assuring Alignment and Safety of Large Language Models (Anwar2024)
Foundationality May Cause Correlated Failures
Foundational Challenges in Assuring Alignment and Safety of Large Language Models (Anwar2024)
Financial instability due to model homogeneity
Risk Sources and Risk Management Measures in Support of Standards for General-Purpose AI Systems (Gipiškis2024)
Impact on Financial Stability
Frontier AI Risk Management Framework (v1.0) (Tse2025)
Multi-agent collaboration capability
Frontier AI Risk Management Framework (v1.0) (Tse2025)
Multi-agent collusion propensity:
Frontier AI Risk Management Framework (v1.0) (Tse2025)

Other entries from Hammond2025

Miscoordination Risk Category · 7.6
Miscoordination Additional evidence
Incompatible strategies Risk Sub-Category · 7.6
Incompatible strategies Additional evidence
Credit Assignment Risk Sub-Category · 7.6
Limited Interactions Risk Sub-Category · 7.6
Conflict Risk Category · 7.6
Conflict Additional evidence
Conflict Additional evidence
Social Dilemmas Risk Sub-Category · 7.6
Military Domains Risk Sub-Category · 7.6
Coercion and Extortion Risk Sub-Category · 7.6