OpenAI Guardrails

# Fix list: OpenAI Guardrails

From Anchor Terminal's listing at https://www.anchorterminal.com/tools/openai-guardrails, the October 2026 research run, assessed 8 October 2026. Grade B, 69.5 out of 100.

This is everything the published grade says the listing lacks, the biggest possible gain to the total first. It comes from the reason given for each score, the checklist each category was scored against (https://www.anchorterminal.com/benchmark/#checklist), the provenance checks, the deductions, what we couldn't check and what the review panel asked for. A fix counts at the next check, once it's public.

For a coding agent working on OpenAI Guardrails: work through the items below in the product, its docs and its public pages. Each category gives the reason for its score, with the points each checklist item earned, and the checklist itself, so the gap is the items that earned less than their points. Change the product, not the wording, and keep a note of what you changed and where it's published.

## 1. Security & auth, 59 out of 100, up to 7.2 more on the total

Why it scored 59: Framework reading. No credential of its own. It uses the OpenAI API key, or the key of whichever compatible endpoint is set, from the environment or the constructor (10 of 30). The prompt injection check runs before and after each tool call in the Agents SDK and can reject or halt, but there is no human approval step, and check failures pass by default (10 of 20). Jailbreak and prompt injection detection are the product, with a published benchmark in which the default model scores 0.000 recall at a 1 per cent false positive rate (13 of 15). Per-call results with check name, stage, confidence and token usage, and no log sink of its own (9 of 15). SECURITY.md points to OpenAI's coordinated disclosure policy, security.txt names a Bugcrowd contact, CodeQL and Dependabot are configured, releases use PyPI trusted publishing with attestations, and the repository has no published advisories (17 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-security):

- 0 to 30, the credential model. 30 for OAuth 2.1 with scopes, or scoped and revocable keys with rotation. 20 for plain revocable API keys. 10 for one all-powerful key. 10 off when a secret can travel in a URL query string as a documented option.
- 0 to 20, read-only or least-privilege modes, and confirmation or approval for destructive actions.
- 0 to 15, prompt-injection posture where the tool returns untrusted content (documented mitigations or guidance). A tool that returns no untrusted content gets 10.
- 0 to 15, audit logs or per-call visibility for the operator.
- 0 to 20, a security programme. security.txt or a disclosure policy, a bug bounty, SOC 2 or ISO 27001, advisories handled in public.

Models are read for retention, whether API data trains models (and whether that's off by default), zero-retention options and certifications. Frameworks for telemetry defaults, approval hooks, guardrails and sandboxing.

## 2. Schema & documentation, 66 out of 100, up to 5.5 more on the total

Why it scored 66: Framework reading. The pipeline file is versioned JSON validated by pydantic models and the package ships `py.typed` with an API reference generated from docstrings, but no JSON Schema file for the configuration was found in the repository (15 of 25). `llms.txt` on the docs site returns 404 (0). Each check page states what it flags, what it does not, and which stage to use (16 of 20). Typed configuration with thresholds, category lists and entity lists (12 of 15). Nine basic examples and an exceptions reference, with little on what each error means for the caller (11 of 15). Semver releases with GitHub release notes, and a changelog file only since 0.3.3 (12 of 15). The docs deploy workflow failed on main on 8 October 2026.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-schema):

APIs and MCP servers.

- 25, a machine-readable contract (a public OpenAPI file or similar; for MCP, typed JSON Schema inputs on every tool).
- 10, llms.txt or Markdown docs served for agents.
- 0 to 20, descriptions that say what a tool is for, when to use it and when not to, read from the tool definitions in the source or the API reference.
- 0 to 15, typed inputs with enums, constraints and required fields, and no free-form JSON blobs.
- 0 to 15, examples and documented error responses.
- 15, versioning and a public changelog.

Models are read from the API reference, the OpenAPI file, llms.txt, the structured-output and tool-use docs and the model cards. Frameworks from docs a model can follow, typed interfaces, examples and the API reference.

## 3. Reliability, 73 out of 100, up to 5.4 more on the total

Why it scored 73: Local-software reading. Installs from PyPI as `openai-guardrails`, Python 3.11 to 3.14 stated (20). Public CI runs ruff, mypy, pyright and the test suite on four Python versions, and the CI run on main passed on 8 October 2026 (25). No open issues and 11 closed in total, though four were closed together on 18 August 2026 after three to ten months (20 of 25). Semver tags, but CHANGELOG.md starts at 0.3.3 and no release note marks a breaking change. RELEASING.md says breaking changes bump the minor version before 1.0 (8 of 15). Version 0.3.3 and the README title reads Preview (0).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-reliability):

Hosted APIs, MCP servers, models and platforms.

- 20, a public status page with component history (Statuspage, Instatus, BetterStack or the vendor's own).
- 0 to 30, the incident record for the last 90 days on that page. 30 for a clean record or trivial incidents only, 20 for minor incidents only, 10 for one major outage (an hour or more of a core API down, or errors across the board), 0 for several. 5 when there's no history we could read, and the note says so.
- 15, rate limits documented with numbers.
- 15, documented 429 or overload handling (Retry-After, backoff guidance), and idempotency keys or safe-retry guidance where writes are involved.
- 10, an SLA published for any paid tier.
- 10, the surface agents use is generally available, not beta or preview.

Local packages, SDKs, frameworks and stdio MCP servers.

- 20, installs from an official package with supported runtimes stated.
- 25, a public CI and test suite, passing on the default branch.
- 0 to 25, open crash or regression issues relative to activity (25 for few and handled, 0 for many, old and unanswered).
- 15, semver discipline and breaking changes called out in a changelog.
- 15, version 1.0 or later, or declared stable.

Protocols are read from their reference implementations, the public facilitators or servers, spec stability and test vectors.

## 4. Payments & pricing, 60 out of 100, up to 5 more on the total

Why it scored 60: Self-hosted rule. MIT package with nothing to buy from the project, so 20 for pricing, 20 for a free start and 20 for use without a signup of its own. No payment protocol (0). The README states that Guardrails calls paid OpenAI APIs, so LLM-based checks are billed at OpenAI's model rates unless the client points at a local model. The moderation check page says that call has no cost.

The checklist (https://www.anchorterminal.com/benchmark/#checklist-payments):

The published rubric, also on the [x402 page](https://www.anchorterminal.com/x402/).

- 40, a machine payment protocol (x402, MPP or L402) on the tool's own endpoints. 10 to 30 when it covers only some endpoints or only goes through a third party, and the note says which.
- 20, per-call or per-unit pricing published without a login. 10 for public plan-only pricing, 0 for "contact sales" or prices behind a login.
- 20, a free tier or trial that doesn't need a card.
- 20, autonomous onboarding, meaning an agent can get access without a person signing up in a browser (keyless use, x402, a programmatic key API).

Payment platforms and agent wallets rarely charge for their own API over a machine protocol, so the first line has steps for them, and the highest one that applies counts. 40 when x402, MPP or L402 runs on all their own endpoints, 30 when it runs on part of their own API, 25 when their merchants can accept one, 20 for running a facilitator, 15 for paying as a buyer, and 0 when the only protocol is their own. Merchant acceptance sits above a facilitator because the platform's own customers can charge agents through it, while a facilitator settles for sellers who wire up the protocol themselves. The counter-argument (a facilitator does more for the protocol as a whole) has a point. Each note says which step applied.

Open-source software you run yourself is scored on its hosted or paid option if it has one. A free, self-hosted package with nothing to buy gets 20, 20 and 20 for the last three lines, and 0 to 40 for the first only if it ships a payment protocol.

## 5. Agent ergonomics, 73 out of 100, up to 4.4 more on the total

Why it scored 73: Framework reading. `GuardrailsOpenAI` replaces `OpenAI` with one config argument, and results sit on `response.guardrail_results` (22 of 25). Confidence thresholds, `max_turns`, `include_reasoning` off by default and `suppress_tripwire` control what comes back, and `total_guardrail_token_usage` reports the cost (15 of 20). Typed exceptions carry the check name and stage, but a check that fails to run is reported as not triggered unless `raise_guardrail_errors=True` (12 of 20). Checks are stateless and safe to repeat, each LLM-based check is an extra model call, and streaming can show output before an output check trips (12 of 20). Python and TypeScript packages, few required parameters, and a spaCy model to install by hand for Contains PII (12 of 15).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-ergonomics):

- 0 to 25, context cost. For MCP, the number and size of the tool definitions (25 for ten or fewer compact tools, 15 for 11 to 30, 5 for more than 30, plus up to 10 back for toolsets, dynamic loading or read-only subsets). For APIs, whether responses can be sized (field selection, limits, summaries).
- 20, pagination, filtering and output-size controls.
- 20, actionable, documented error responses, codes and messages an agent can recover from.
- 20, idempotency or safe retries, and for MCP the `readOnlyHint` and `destructiveHint` annotations.
- 15, sensible defaults, few required parameters, and official SDKs in at least two languages.

Models are read for tool use, structured output, prompt caching, context length, batch and SDKs. Frameworks for how much code and how many defaults a tool-calling agent with MCP needs.

## 6. Transparency & trust, 76 out of 100, up to 2.1 more on the total

Made of editorial 65, provenance 87.

Why it scored 76: MIT licence in the repository and on PyPI (30). The README says the developer is responsible for storage of blocked content and that the library calls OpenAI APIs, and each check page says whether it uses a model, but no page lists which checks send text off the machine, and OpenAI's API terms and privacy policy answered 403 to us today so were not read (15 of 30). No deprecation policy. RELEASING.md states that breaking changes bump the minor version before 1.0 (8 of 20). No telemetry was found in the source. Requests to api.openai.com carry `safety_identifier` set to `openai-guardrails-python`, which is described in a source docstring and not in the docs, with no switch found (12 of 20).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-transparency):

- 0 to 30, source availability and licence clarity. 30 for open source under an OSI licence, 15 for closed with clear terms, 0 for unclear terms.
- 0 to 30, data handling and retention statements that agree with each other (privacy policy, DPA, retention periods, subprocessors).
- 0 to 20, a deprecation policy or notices with dates.
- 0 to 20, telemetry disclosed with an opt-out (local software), or subprocessors and data locations disclosed (hosted).

The other half of Transparency and trust is the provenance score, computed from checked facts (below). The category score is the mean of the two.

Provenance checks not met in full (half of this category, computed from checked facts):

- Status page: not found (0 of 10)

## 7. Maintenance & community, 89 out of 100, up to 1 more on the total

Why it scored 89: 0.3.3 on 10 September 2026, 28 days before the check (30). Three PyPI releases in the last 90 days, 0.3.0 on 21 July, 0.3.2 on 21 August and 0.3.3 (20). All 11 issues are closed and bug reports from 2025 were answered within days, but four were closed in bulk on 18 August 2026, outside pull requests are refused by policy, and nothing was released from 15 December 2025 to 21 July 2026 (15 of 25). Current official packages on PyPI and npm (15). Dependabot, CodeQL and CI on four Python versions, with the docs deploy failing on 8 October (9 of 10).

The checklist (https://www.anchorterminal.com/benchmark/#checklist-maintenance):

- 0 to 30, time since the last release, or the last published model or API change for a closed service. 30 within 30 days, 20 within 90, 10 within 180, 0 older.
- 20, at least three releases or dated changelog entries in the last 90 days.
- 0 to 25, responsiveness. Issues and pull requests answered on GitHub (the open issues and how recent the replies are). For closed services, a public changelog and a support or community channel that answers, 0 to 15.
- 15, presence in the official MCP registry under a verified namespace (MCP servers), or current official SDKs (APIs and models).
- 10, package health, current dependencies and CI.

Models are read for deprecation notice periods and model churn rather than release counts.

## What we couldn't check

What we couldn't read counted as absent. Publishing it on a page a plain HTTP fetch can read (not only in a browser) lets the next check count it.

- unchecked: https://openai.com/policies/services-agreement/ and https://openai.com/policies/privacy-policy/ answered HTTP 403 to our reader on 8 October 2026, so the terms and privacy policy that cover the OpenAI API calls the checks make were not read and `provenance.terms` and `provenance.privacy` are left out.
- unchecked: https://openai.com/security/disclosure/ answered HTTP 403, so the disclosure policy and any bounty scope for this package were not read.
- unchecked: https://guardrails.openai.com/ is drawn by script, so the configuration wizard and whatever terms it links were not read.
- Whether `safety_identifier` can be turned off for calls to api.openai.com. No switch was found in the source.
- Why no release was published between 15 December 2025 and 21 July 2026.
- The tag v0.3.1 exists in the repository and no 0.3.1 is on PyPI.

## Weaknesses

- By default a check that fails to run returns `tripwire_triggered=False`, so the request continues. Strict mode is opt-in
- The README titles the package a preview, the version is 0.3.3, and no release was published between 15 December 2025 and 21 July 2026
- With `stream=True` the output checks run alongside the stream, and the docs say violating content may appear briefly
- The docs' own table gives the default jailbreak model, `gpt-4.1-mini`, a recall of 0.000 at a 1 per cent false positive rate
- LLM-based checks add a billed model call each. The docs list a median of 1,538 ms for `gpt-4.1-mini` on the jailbreak check
- Pull requests from non-collaborators are not accepted, and CHANGELOG.md starts at 0.3.3

## What costs an agent a turn today

The notes we give agents before they call it. Each one is a workaround an agent shouldn't need.

- Pass `raise_guardrail_errors=True` to the client. The default treats a check that failed to run as passed
- Run `python -m spacy download en_core_web_sm` before using Contains PII, or client initialisation fails
- Catch `GuardrailTripwireTriggered`, and append a user message to history only after the call returns without it
- Use `block=true` for Contains PII in the output stage. Masking works only in the pre-flight stage
- Keep `stream=False` where output must be checked before it is shown, and budget one extra model call per LLM-based check

## When it's done

Send what changed and where it's published as a dispute (https://www.anchorterminal.com/builders/#disputes, or `POST https://www.anchorterminal.com/api/v1/contact` with `"kind": "dispute"`). Disputes are answered in public, and the listing is checked again by the same checklist. Paying for an audit or a listing claim changes nothing here.