aicordon-haystack 0.1.0__py3-none-any.whl

aicordon_haystack-0.1.0.dist-info/METADATA

@@ -0,0 +1,199 @@
+Metadata-Version: 2.5
+Name: aicordon-haystack
+Version: 0.1.0
+Summary: Check what an LLM is given for prompt injection: material at ingest, and the turn it answers
+Project-URL: Homepage, https://github.com/AICordon/aicordon/blob/main/integrations/haystack/README.md
+Project-URL: Repository, https://github.com/AICordon/aicordon
+Project-URL: Issues, https://github.com/AICordon/aicordon/issues
+Project-URL: Changelog, https://github.com/AICordon/aicordon/blob/main/integrations/haystack/CHANGELOG.md
+Author-email: Mikhail Gribov <mihail.gribov.rs@gmail.com>
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: guardrails,haystack,jailbreak,llm-security,prompt-injection,rag
+Classifier: Development Status :: 4 - Beta
+Classifier: Intended Audience :: Developers
+Classifier: Programming Language :: Python :: 3
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Classifier: Topic :: Security
+Requires-Python: >=3.10
+Requires-Dist: aicordon>=1.1.0
+Requires-Dist: haystack-ai>=3.0.0
+Description-Content-Type: text/markdown
+
+# AI Cordon Picket for Haystack
+
+Check what an LLM is given for prompt injection — in both places it can arrive:
+
+| component | reads | with |
+|---|---|---|
+| `PromptInjectionFilter` | **material**: documents at ingest, before they are chunked and embedded | Picket's `ipi` rules |
+| `PromptInjectionGuard` | **the request**: the turn the model is about to answer | Picket's `dpi` rules |
+
+The two rule sets are disjoint, and neither is a stricter version of the other — this is not a
+sensitivity knob. Pick by role: material is what the model works on, the request is what it answers.
+Your code knows which is which; it puts them in different places when it assembles the call.
+
+The check is a rule, not a model: no GPU, no network, no key, a few hundred kilobytes of base, and a
+fraction of a millisecond per turn on one core — see [what it costs](#what-it-costs).
+
+## Installation
+
+```bash
+pip install aicordon-haystack
+```
+
+## Material at ingest
+
+```python
+from haystack import Pipeline
+from haystack.components.preprocessors import DocumentSplitter
+from haystack.components.writers import DocumentWriter
+from haystack_integrations.components.preprocessors.aicordon import PromptInjectionFilter
+
+pipe = Pipeline()
+pipe.add_component("ipi_filter", PromptInjectionFilter(mode="redact"))
+pipe.add_component("splitter", DocumentSplitter(split_by="word", split_length=200))
+pipe.add_component("writer", DocumentWriter(document_store=store))
+
+pipe.connect("converter.documents", "ipi_filter.documents")
+pipe.connect("ipi_filter.documents", "splitter.documents")
+pipe.connect("splitter.documents", "writer.documents")
+pipe.connect("ipi_filter.rejected", "quarantine.documents")   # optional; nothing disappears quietly
+```
+
+The component sits **before the splitter**: a cut here takes the injection out of the chunks, the
+embeddings and the store at once, with no offsets to reconcile across chunk boundaries.
+
+### What it does with a finding
+
+| `mode` | the document | the length |
+|---|---|---|
+| `annotate` | indexed unchanged, the finding recorded in metadata | unchanged |
+| `blank` | every character of the block becomes `blank_char` (default `*`) | **preserved** |
+| `mask` | the block is replaced by `mask_with` | changes |
+| `redact` *(default)* | the block is cut out | changes |
+| `drop` | not indexed; it comes out of the `rejected` socket | — |
+| `fail` | the run stops on the first finding | — |
+
+`blank` is for pipelines that carry offsets, page maps or diffs downstream and cannot have a
+document change length under them.
+
+The cut takes **the line holding the span**, or the sentence when that line runs past 1500
+characters: the span points at the injection, but what must leave the index is the whole utterance.
+Measured on 1200 documents: the payload is gone entirely in 91% of catches, at a median 11.6% of the
+document removed.
+
+## The turn the model answers
+
+```python
+from haystack_integrations.components.validators.aicordon import PromptInjectionGuard
+
+pipe.add_component("guard", PromptInjectionGuard())          # mode="drop" is the default
+pipe.connect("prompt.messages", "guard.messages")
+pipe.connect("guard.messages", "llm.messages")               # the model is called on this path
+pipe.connect("guard.blocked", "refusal.messages")            # and not on this one
+```
+
+**Two sockets, one value.** On a flagged exchange `run` returns `blocked` and no `messages` key, so
+the generator is not called at all. Connect `blocked` to whatever answers the user instead.
+
+The decision is for the **exchange**, not for one message: drop the offending turn and the model
+answers the one before it.
+
+### What it reads, and what it does with a finding
+
+`roles` maps a role to a rule set, default `{"user": "dpi"}`. `assistant` is the model's own text,
+`system` the operator's. `tool` carries material and switches on with
+`roles={"user": "dpi", "tool": "ipi"}` — measure your own tool output first: over live chat text the
+`ipi` rules raise eight times as many alarms as over documents, and they fire on command lists and
+code, which is what a tool result looks like.
+
+| `mode` | the exchange |
+|---|---|
+| `drop` *(default)* | routed to the `blocked` socket; the model is not called |
+| `annotate` | passed through, with the finding in each read message's metadata |
+| `fail` | the run stops with `InjectionFound` |
+
+No mode edits a turn, and asking for one raises. The cut above is fitted to an instruction spliced
+into a document; a typed jailbreak is not spliced into anything — it *is* the turn, and cutting it
+leaves the rest of the attack in place.
+
+## What lands in the metadata
+
+Written on **every** document and every message whose role is read, so "checked and clean" is
+distinguishable from "never checked". A role outside the map gets no fields — a third, distinct
+fact.
+
+```python
+{"ipi_flagged": False, "ipi_action": "none", "ipi_base": "20260817"}
+{"ipi_flagged": True,  "ipi_action": "redact", "ipi_base": "20260817",
+ "ipi_threats": ["IPI/Secret.Reveal.B"], "ipi_spans": [[812, 947]], "ipi_removed_chars": 163}
+
+{"picket_flagged": True, "picket_action": "drop", "picket_base": "20260817",
+ "picket_threats": ["DPI/Policy.Cancel.M"], "picket_spans": [[0, 41]]}
+```
+
+Findings are also logged through Haystack's own logger at `warning`, so the level is yours to set.
+
+## Measured
+
+Not the detector's recall — that ships with the detector — but what the pipeline delivers with the
+component and without.
+
+**Material.** [Quadrat-IPI v1.0.1](https://huggingface.co/datasets/mihailgribov/quadrat-ipi),
+1000 injected and 1000 clean documents, `mode="redact"`; how much of a planted payload still reaches
+the store:
+
+| | whole corpus | injections that ask the model to **reveal** something |
+|---|---|---|
+| payload reaches the store intact, without the filter | 100% | 100% |
+| payload reaches the store intact, with it | **85.4%** | **42.8%** |
+| payload gone without a trace | 13.1% | **52.3%** |
+| clean documents dropped or trimmed | 0 of 1000 | 0 of 1000 |
+
+Both columns matter: the first is an arbitrary stream, the second is where the rule is strong. A
+document cost 8.3 ms in that run — documents are long, and cost follows length.
+
+**The request.** Held-out forum jailbreaks from
+[TrustAIRLab in-the-wild](https://huggingface.co/datasets/TrustAIRLab/in-the-wild-jailbreak-prompts)
+(537, near-duplicates of the fitting half removed) against 20 000 real user turns from
+[WildChat](https://huggingface.co/datasets/allenai/WildChat-1M), `mode="drop"`:
+
+| | |
+|---|---|
+| attacks reaching the model, without the guard | 100% (537 of 537) |
+| attacks reaching the model, with it | **65.2%** |
+| turns not answered, out of 20 000 real ones | 0.070% (14) |
+| verdicts differing from the bare detector | **0** |
+
+WildChat carries no attack labels and real jailbreaks sit inside it, so "turns not answered" is an
+upper bound on the cost to a real user, not a false-alarm rate. The detector's working point, on a
+labelled pool, is in its report.
+
+Reproduce both with `eval/measure.py` and `eval/measure_dialog.py`.
+
+## What it costs
+
+Adding the component to a pipeline costs **0.39 ms for a turn of median length**, and 1.65 ± 0.03 ms
+averaged over ordinary traffic — 3000 real WildChat turns, each timed five times
+(`eval/costturn.py`). Loading the base costs 15 ms, once per process.
+
+The average is four times the median: cost follows turn length, and a chat pool has a long tail.
+Find your row:
+
+| turn length | turns in the pool | cost |
+|---|---|---|
+| under 200 characters | 1943 | 0.30 ms |
+| 200–500 | 411 | 0.74 ms |
+| 500–1500 | 321 | 1.62 ms |
+| 1500–4000 | 182 | 4.09 ms |
+| over 4000 | 143 | 13.20 ms |
+
+Cost drifts with machine load. These were taken in one run by one procedure — the only way two
+figures compare.
+
+No findings does not mean no injection.
+
+## License
+
+Apache-2.0, the same as the detector it wraps.

aicordon_haystack-0.1.0.dist-info/RECORD

@@ -0,0 +1,8 @@
+haystack_integrations/components/preprocessors/aicordon/__init__.py,sha256=Lf4jfiANubNCYu5lgRZbEx7KxE5hUPSBhWgTlFn12Eg,96
+haystack_integrations/components/preprocessors/aicordon/prompt_injection_filter.py,sha256=gbqAB0sGhaxIQJcPMIDdKVfGgZp4G0nr40FOBNi5deE,4179
+haystack_integrations/components/validators/aicordon/__init__.py,sha256=hwa4zx7Q-IzxaBtpAZsukCpHxkrRwmNBQ4TOhjkSb2s,93
+haystack_integrations/components/validators/aicordon/prompt_injection_guard.py,sha256=_7yj6zgDCXAI4Vm-IeBgBEGTGR03YtftZBjHKmv1KIc,6416
+aicordon_haystack-0.1.0.dist-info/METADATA,sha256=RNMwEI_b_HtlpNwuofx4JaV61zd3Ny4fzdAqydBKkZM,8938
+aicordon_haystack-0.1.0.dist-info/WHEEL,sha256=zOwg4jB6zX2kU910N-cMawjivD6tO8NEWvE12je1bVk,87
+aicordon_haystack-0.1.0.dist-info/licenses/LICENSE,sha256=uEmWZek7CjsCTJkApsWk4Oq8XUDmvdCp1MIiiJq_xG8,11345
+aicordon_haystack-0.1.0.dist-info/RECORD,,

aicordon_haystack-0.1.0.dist-info/WHEEL

@@ -0,0 +1,4 @@
+Wheel-Version: 1.0
+Generator: hatchling 1.32.0
+Root-Is-Purelib: true
+Tag: py3-none-any

aicordon_haystack-0.1.0.dist-info/licenses/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 Mikhail Gribov
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

haystack_integrations/components/preprocessors/aicordon/__init__.py

@@ -0,0 +1,3 @@
+from .prompt_injection_filter import PromptInjectionFilter
+
+__all__ = ["PromptInjectionFilter"]

haystack_integrations/components/preprocessors/aicordon/prompt_injection_filter.py

@@ -0,0 +1,78 @@
+"""Check documents for indirect prompt injection before they are chunked and embedded.
+
+The component sits between the converter and the splitter of an indexing pipeline. Everything it
+decides comes from `aicordon.guard.InjectionGuard`; what lives here is the translation into
+Haystack's types and its contract — nothing else, so the same policy serves the other frameworks
+unchanged.
+
+    pipe.add_component("ipi_filter", PromptInjectionFilter(mode="redact"))
+    pipe.connect("converter.documents", "ipi_filter.documents")
+    pipe.connect("ipi_filter.documents", "splitter.documents")
+    pipe.connect("ipi_filter.rejected", "quarantine.documents")   # optional, nothing is lost silently
+"""
+from __future__ import annotations
+
+from dataclasses import replace
+from typing import Any
+
+from aicordon.guard import InjectionGuard
+from haystack import Document, component, default_from_dict, default_to_dict, logging
+
+logger = logging.getLogger(__name__)
+
+
+@component
+class PromptInjectionFilter:
+    """Two outputs on purpose: what goes on to be indexed, and what was taken out of the stream.
+
+    A component that only returned the survivors would drop documents where the pipeline diagram
+    shows a single arrow, and a reader of that diagram would never learn it. With `rejected` as its
+    own socket, removal is a connection someone chose to make — to a quarantine store, to a log, or
+    to nothing at all, but visibly.
+    """
+
+    def __init__(self, mode: str = "redact", meta_prefix: str = "ipi", blank_char: str = "*",
+                 mask_with: str = "[prompt injection removed]") -> None:
+        # Only strings here: Haystack requires init parameters to be JSON-serialisable so that a
+        # pipeline can be saved and loaded. The detector is built in `warm_up()` instead, which is
+        # the hook the framework offers for state too heavy to raise during pipeline validation.
+        #
+        # KEPT ON THE INSTANCE UNDER THE PARAMETER NAMES, and serialised explicitly below. Without
+        # both, saving a pipeline loses the settings SILENTLY: with no `to_dict` the framework reads
+        # each init parameter back with `getattr(self, name)`, and when that fails it falls back to
+        # the default in the signature. A pipeline built with mode="annotate" came back out of YAML
+        # as "redact", with nothing raised anywhere.
+        self.mode = mode
+        self.meta_prefix = meta_prefix
+        self.blank_char = blank_char
+        self.mask_with = mask_with
+        self._guard = InjectionGuard(mode=mode, meta_prefix=meta_prefix, blank_char=blank_char,
+                                     mask_with=mask_with)
+
+    def to_dict(self) -> dict[str, Any]:
+        return default_to_dict(self, mode=self.mode, meta_prefix=self.meta_prefix,
+                               blank_char=self.blank_char, mask_with=self.mask_with)
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "PromptInjectionFilter":
+        return default_from_dict(cls, data)
+
+    def warm_up(self) -> None:
+        self._guard.warm_up()
+
+    @component.output_types(documents=list[Document], rejected=list[Document])
+    def run(self, documents: list[Document]) -> dict[str, Any]:
+        kept: list[Document] = []
+        rejected: list[Document] = []
+        for doc in documents:
+            verdict = self._guard.inspect(doc.content or "")
+            if verdict.flagged:
+                # Logged through the host's logger, at a level the host controls: "write it to the
+                # log" is not a mode — it is wanted under `drop` and under `annotate` alike.
+                logger.warning("prompt injection in a document at ingest: {threats}, action {mode}",
+                               threats=", ".join(verdict.threats), mode=self.mode)
+            # A copy, never the input: the same list can be connected to a second branch of the
+            # pipeline, and editing in place would rewrite that branch's documents too.
+            out = replace(doc, content=verdict.text, meta={**doc.meta, **self._guard.meta(verdict)})
+            (kept if verdict.keep else rejected).append(out)
+        return {"documents": kept, "rejected": rejected}

haystack_integrations/components/validators/aicordon/__init__.py

@@ -0,0 +1,3 @@
+from .prompt_injection_guard import PromptInjectionGuard
+
+__all__ = ["PromptInjectionGuard"]

haystack_integrations/components/validators/aicordon/prompt_injection_guard.py

@@ -0,0 +1,114 @@
+"""Check the exchange for a jailbreak before the model is called.
+
+The component sits between whatever assembles the messages and the chat generator. Everything it
+decides comes from `aicordon.guard.DialogueGuard`; what lives here is the translation into
+Haystack's types and its contract — nothing else, so the same policy serves the other frameworks
+unchanged.
+
+    pipe.add_component("guard", PromptInjectionGuard())          # mode="drop" by default
+    pipe.connect("prompt.messages", "guard.messages")
+    pipe.connect("guard.messages", "llm.messages")               # the model is called on this path
+    pipe.connect("guard.blocked", "refusal.messages")            # and not on this one
+
+TWO SOCKETS, AND ONLY ONE OF THEM CARRIES A VALUE. On a flagged exchange `run` returns `blocked` and
+no `messages` key at all, which is how a Haystack component branches: a receiver whose socket got
+nothing does not run. So the generator is not called with a mutilated message list — it is not
+called. Whoever wants a reply for the user connects `blocked` to something that produces one.
+
+THE COMPANION IS `PromptInjectionFilter`, and they are not variants of each other. That one reads
+material at ingest with the `ipi` rules and may rewrite it; this one reads the request with the
+`dpi` rules and never does. Which applies is decided by the role a string plays in the prompt — the
+module docstrings in `aicordon.guard` set out why, and why nothing here cuts a hole in a turn.
+"""
+from __future__ import annotations
+
+from dataclasses import replace
+from typing import Any
+
+from aicordon.guard import DEFAULT_ROLES, DialogueGuard
+from haystack import component, default_from_dict, default_to_dict, logging
+from haystack.dataclasses import ChatMessage
+
+logger = logging.getLogger(__name__)
+
+#: What the guard reads of a message. A message can carry several text parts alongside images and
+#: files; the parts are joined because that is what the model is given, and the non-text content is
+#: not read at all — Picket is a rule over text.
+PART_SEPARATOR = "\n\n"
+
+
+@component
+class PromptInjectionGuard:
+    """Decides for the exchange as a whole, not message by message.
+
+    :param mode: `drop` routes a flagged exchange to `blocked` instead of to the model; `annotate`
+        lets everything through with the finding in each read message's metadata, for a prompt or a
+        downstream router to act on; `fail` raises `InjectionFound`.
+    :param roles: role name to rule set — `dpi` for a typed request, `ipi` for material. Defaults to
+        `{"user": "dpi"}`. `tool` is left out on purpose: a tool result is material, but it reads
+        like agent prompts and code, where the `ipi` rules raise eight times the alarms they raise
+        on documents. Switch it on and measure your own tool outputs first.
+    :param meta_prefix: prefix for the metadata keys written on each message that was read.
+    """
+
+    def __init__(self, mode: str = "drop", roles: dict[str, str] | None = None,
+                 meta_prefix: str = "picket") -> None:
+        # Only strings and plain dicts here: Haystack requires init parameters to be
+        # JSON-serialisable so that a pipeline can be saved and loaded. The detector is built in
+        # `warm_up()` instead, which is the hook the framework offers for state too heavy to raise
+        # during pipeline validation.
+        #
+        # KEPT ON THE INSTANCE UNDER THE PARAMETER NAMES, and serialised explicitly below. Without
+        # both, saving a pipeline loses the settings SILENTLY: with no `to_dict` the framework reads
+        # each init parameter back with `getattr(self, name)`, and when that fails it falls back to
+        # the default in the signature.
+        self.mode = mode
+        self.roles = dict(DEFAULT_ROLES if roles is None else roles)
+        self.meta_prefix = meta_prefix
+        self._guard = DialogueGuard(mode=mode, roles=self.roles, meta_prefix=meta_prefix)
+
+    def to_dict(self) -> dict[str, Any]:
+        return default_to_dict(self, mode=self.mode, roles=self.roles,
+                               meta_prefix=self.meta_prefix)
+
+    @classmethod
+    def from_dict(cls, data: dict[str, Any]) -> "PromptInjectionGuard":
+        return default_from_dict(cls, data)
+
+    def warm_up(self) -> None:
+        self._guard.warm_up()
+
+    @component.output_types(messages=list[ChatMessage], blocked=list[ChatMessage])
+    def run(self, messages: list[ChatMessage]) -> dict[str, Any]:
+        verdict = self._guard.decide([(m.role.value, self._text(m)) for m in messages])
+        if verdict.flagged:
+            # Logged through the host's logger, at a level the host controls: "write it to the log"
+            # is not a mode — it is wanted under `drop` and under `annotate` alike.
+            logger.warning("prompt injection in a turn: {threats}, action {mode}",
+                           threats=", ".join(verdict.threats), mode=self.mode)
+        # Copies, never the inputs: the same list can be connected to a second branch of the
+        # pipeline, and editing in place would rewrite that branch's messages too. Haystack warns
+        # on in-place mutation of a ChatMessage for exactly this reason.
+        out = [self._with_meta(m, self._guard.meta(verdict, i)) for i, m in enumerate(messages)]
+        # One key, not two. The socket that is absent is the branch that does not run.
+        return {"messages": out} if verdict.keep else {"blocked": out}
+
+    @staticmethod
+    def _text(message: ChatMessage) -> str:
+        """Everything of the message that is text the model will read.
+
+        `texts` alone is not that. A tool message carries its content in `tool_call_result.result`
+        and its `texts` is EMPTY, so a guard reading only `texts` would check tool results by
+        returning "nothing found" on an empty string — the quiet kind of wrong, since the role is in
+        the map and the metadata says it was read. Tool CALLS are left out on purpose: those are the
+        model's own output, not something handed to it.
+        """
+        parts = list(message.texts)
+        parts += [r.result for r in message.tool_call_results if r.result]
+        return PART_SEPARATOR.join(parts)
+
+    @staticmethod
+    def _with_meta(message: ChatMessage, extra: dict[str, Any]) -> ChatMessage:
+        if not extra:
+            return message
+        return replace(message, _meta={**message.meta, **extra})