Protecting User Prompts Via Character-Level Differential Privacy

Protecting User Prompts Via Character-Level Differential Privacy | Signal Canvas | ScienceToStartup

Page Freshness

Signal Canvas proof surface

Canonical route: /signal-canvas/protecting-user-prompts-via-character-level-differential-privacy

stale

Proof freshness: stale
Proof status: unverified
Display score: 7/10
Last proof check: 2026-03-30
Score updated: 2026-04-02
Score fresh until: 2026-05-02
References: 11
Source count: 5
Coverage: 50%

This page is showing the last landed evidence receipt and score bundle because the latest proof data is outside the freshness window.

Agent Handoff

Canonical ID protecting-user-prompts-via-character-level-differential-privacy | Route /signal-canvas/protecting-user-prompts-via-character-level-differential-privacy

REST example

curl https://sciencetostartup.com/api/v1/agent-handoff/signal-canvas/protecting-user-prompts-via-character-level-differential-privacy

MCP example

{
  "tool": "search_signal_canvas",
  "arguments": {
    "mode": "paper",
    "paper_ref": "protecting-user-prompts-via-character-level-differential-privacy",
    "query_text": "Summarize Protecting User Prompts Via Character-Level Differential Privacy"
  }
}

source_context

{
  "surface": "signal_canvas",
  "mode": "paper",
  "query": "Protecting User Prompts Via Character-Level Differential Privacy",
  "normalized_query": "2603.26032",
  "route": "/signal-canvas/protecting-user-prompts-via-character-level-differential-privacy",
  "paper_ref": "protecting-user-prompts-via-character-level-differential-privacy",
  "topic_slug": null,
  "benchmark_ref": null,
  "dataset_ref": null
}

Evidence Receipt

Route status: building

Claims: 12

References: 11

Proof: Verification pending

Freshness state: computing

Source paper: Protecting User Prompts Via Character-Level Differential Privacy

PDF: https://arxiv.org/pdf/2603.26032v1

Source count: 5

Coverage: 50%

Last proof check: 2026-03-30T21:55:29.346Z

Signal Canvas receipt window

Watch and verify: Protecting User Prompts Via Character-Level Differential Privacy

/buildability/protecting-user-prompts-via-character-level-differential-privacy

Watchwatch

Subject: Protecting User Prompts Via Character-Level Differential Privacy

Verdict

Watch

Verdict is Watch because viability or proof quality is intermediate and should be re-evaluated before execution.

Preparing verified analysis

GitHub Code Pulse

No public code linked for this paper yet.

Claim map

Strong 12Mixed 0Weak 0

Evidencepartial
In this work, we propose a new method to sanitize user prompts. Our mechanism uses the randomized response mechanism of differential privacy to randomly and independently perturb each character in a word.
Implicationpartial
This is the core method described in the abstract and title.
Verificationpartial
partial
Evidencepartial
Our mechanism uses the randomized response mechanism of differential privacy to randomly and independently perturb each character in a word.
Implicationpartial
This detail about the perturbation mechanism is explicitly stated in the abstract.
Verificationpartial
partial
Evidencepartial
The perturbed text is then sent to a remote LLM, which first performs a prompt restoration and subsequently performs the intended downstream task.
Implicationpartial
The abstract clearly outlines the workflow involving the LLM's restoration step.
Verificationpartial
partial
Evidencepartial
Our results show that sensitive PII tagged in these datasets are reconstructed at a rate close to the theoretical rate of reconstructing completely random words, whereas non-sensitive words are reconstructed at a much higher rate.
Implicationpartial
This is a key experimental result presented in the abstract, indicating effective privacy for sensitive data.
Verificationpartial
partial
Evidencepartial
Our results show that sensitive PII tagged in these datasets are reconstructed at a rate close to the theoretical rate of reconstructing completely random words, whereas non-sensitive words are reconstructed at a much higher rate.
Implicationpartial
This is a direct comparison presented in the results, highlighting the differential reconstruction rates.
Verificationpartial
partial
Evidencepartial
Our method has the advantage that it can be applied without explicitly identifying sensitive pieces of information in the prompt, while showing a good privacy-utility tradeoff for downstream tasks.
Implicationpartial
This is presented as a significant advantage of the proposed method in the abstract.
Verificationpartial
partial
Evidencepartial
Our method has the advantage that it can be applied without explicitly identifying sensitive pieces of information in the prompt, while showing a good privacy-utility tradeoff for downstream tasks.
Implicationpartial
This is a summary statement of the method's performance and benefit.
Verificationpartial
partial
Evidencepartial
The advantage of our mechanism is that we automatically provide privacy while maintaining utility of the prompt, without the need to explicitly identify or mark sensitive terms, using predefined rules or external classifiers, and thus avoiding the pitfalls of incomplete redaction, as may be the case with NER-based schemes.
Implicationpartial
This is stated as a benefit compared to existing methods.
Verificationpartial
partial
Evidencepartial
Our mechanism uses the randomized response mechanism of differential privacy to randomly and independently perturb each character in a word.
Implicationpartial
This is a core description of the proposed method, explicitly stated in the abstract and background.
Verificationpartial
partial
Evidencepartial
The idea is that the restoration will be able to reconstruct non-sensitive words even when they are perturbed due to cues from the context, as well as the fact that these words are often very common. On the other hand, perturbation would make reconstruction of sensitive words difficult because they are rare.
Implicationpartial
This explains the underlying principle and expected outcome of the proposed method, as detailed in the abstract.
Verificationpartial
partial
Evidencepartial
Our results show that sensitive PII tagged in these datasets are reconstructed at a rate close to the theoretical rate of reconstructing completely random words, whereas non-sensitive words are reconstructed at a much higher rate.
Implicationpartial
This is a key experimental result presented in the abstract and elaborated in the background section.
Verificationpartial
partial
Evidencepartial
Our method has the advantage that it can be applied without explicitly identifying sensitive pieces of information in the prompt, while showing a good privacy-utility tradeoff for downstream tasks.
Implicationpartial
This highlights a significant advantage of the proposed method over existing techniques, as stated in the abstract and background.
Verificationpartial
partial

Author intelligence and commercialization panels stay hidden until the proof receipt is verified, cites at least 3 references, includes at least 2 sources, and clears 50% coverage. The paper narrative and citation surfaces remain public while verification is pending.

Protecting User Prompts Via Character-Level Differential Privacy

Use Signal Canvas as the narrative proof surface

Use this Signal Canvas via API or MCP

Signal Canvas proof surface