Invisible Prompt Injection Inside Encrypted Reasoning

Part I

Problem Definition

Encrypted, encoded, hidden, or dynamically assembled content can carry instructions that are invisible to a human reviewer but become executable context when an agent decodes or consumes it [R1] [R2].

This note treats invisible prompt injection inside encrypted reasoning as a bounded control question. It does not infer universal product behavior from a paper, standard, interface screenshot, or single test. A demonstrated result applies only to the cited configuration; an authoritative source defines a requirement or design direction but does not certify an implementation. ACE therefore asks whether the tested system can produce the required behavior and evidence under declared versions, policies, topology, identities, and failure conditions.

The operational distinction is between a claim and a control that can be re-performed. The positive control is: An encrypted task payload decodes to authorized data and produces only a policy-approved action. The adversarial condition is: A hidden or encoded instruction materializes after initial scanning and asks the agent to exfiltrate data through an available tool. A useful result must show both that authorized work remains possible and that the prohibited path is stopped before an irreversible side effect. Missing fields are classified as insufficient evidence, not silently converted into a pass.

Real-world impact

• Encryption can hide malicious instructions from the same reviewers and scanners it is expected to reassure.
• A static scan can pass before a browser or tool dynamically assembles the actual prompt.
• A control claim without versioned evidence can mislead procurement, audit, incident response, and system owners.
• Failing closed without a positive control can conceal a denial-of-service design rather than demonstrate trustworthy behavior.

Part II

Mitigation Direction

Research-backed direction

1. 01
   research proposed

   Verify permissions attached to protected prompts before any derived action executes [R1].

2. 02
   research proposed

   Keep untrusted external data outside the model's instruction context where the task can be executed in an isolated processor [R2].

3. 03
   research proposed

   Detect runtime-assembled and hidden document instructions after materialization, not only in static source [R3].

LogionOS engineering mapping

Implementation hypotheses only. No production validation is claimed.

1. 01
   implementation hypothesis

   Add a versioned policy object for invisible prompt injection inside encrypted reasoning and keep its decision inputs outside model-writable context.

2. 02
   implementation hypothesis

   Generate a signed technical receipt linking actor, request, policy version, decision, enforcement point, and observed outcome.

3. 03
   implementation hypothesis

   Export missing evidence explicitly as insufficient evidence and limit every claim to the tested configuration.

Part III

Consequences and Research Agenda

Consequences

• Encryption can hide malicious instructions from the same reviewers and scanners it is expected to reassure.
• A static scan can pass before a browser or tool dynamically assembles the actual prompt.
• A failed acceptance test requires the related capability claim to remain not demonstrated or insufficient evidence.
• A passing test supports only the declared configuration and does not establish universal safety.

Second-order effects

• Stronger enforcement can increase latency, state, operational dependencies, and legitimate denials.
• More evidence can increase privacy and retention exposure unless raw content is minimized and access-controlled.
• A detector or policy service can become a new failure point and must have explicit fail behavior.
• Attackers may adapt to published checks, so the public test direction should be paired with private regression variants.

Limitations

• Several cited AI-agent and reasoning-security sources are preprints or bounded experiments; they are identified as such in the references.
• The proposed ACE acceptance test has not yet been run across all incumbent and AI-native implementations.
• Cryptographic integrity proves that an artifact was not altered after commitment; it does not prove that the artifact was true, complete, or correctly interpreted.
• Legal and contractual applicability remains deployment- and jurisdiction-specific.

Open research questions

1. Which transformations must trigger renewed inspection and authorization?
2. When is execute-only processing practical without losing necessary model utility?
3. Which evidence fields are mandatory for a demonstrated result, and which may be not applicable?
4. How should continuous regression detect policy, model, tool, and provider drift after the initial test?

Sources

References

1. [R1]
   Encrypted Prompt: Securing LLM Applications Against Unauthorized Actions

   preprint · original-paper

2. [R2]
   Execute-Only Agents: Architectural Defense Against Prompt Injection for AI Agents

   workshop-paper · original-paper

3. [R3]
   Fooling AI Agents: Web-Based Indirect Prompt Injection Observed in the Wild

   published · primary-technical-source

Record

Publication Record

Recommended citation

Ma, Chris. “Invisible Prompt Injection Inside Encrypted Reasoning.” ACE Research Note ACE-RN-2026-018, v1.0, 2026.

Corrections

No corrections recorded.

Organizational disclosure

ACE Research and LogionOS share organizational affiliation. LogionOS mappings in this note are implementation hypotheses, not independently validated product claims.

Evidence boundary

This note synthesizes cited public research and defines an ACE acceptance direction. It does not report a completed cross-vendor experiment unless explicitly stated, and it contains no private ACE prompts, holdout identifiers, customer data, or raw model responses.