llmaudit.eu

LLM security audit · European Union

Scope an audit

Independent LLM security audit

Find out what your chatbot, RAG or agent can be talked into doing

An LLM security audit tests the application around the model: the prompt, the documents it retrieves, the tools it can call, the output it renders and the logs it leaves behind. This page sets out what is tested, what the report has to contain, and which EU AI Act duties that evidence answers.

Written by the OffSeq security team. The method, the report structure and the article mapping below are the ones used on paid engagements; reading them costs nothing.

§ 01 Definition

What an LLM security audit is, and what it is not

The phrase is used for at least four different things. On this site it means one: a security assessment of a deployed LLM application, carried out adversarially, with findings that can be reproduced from the transcript.

In scope

  • The instruction boundary. Whether user text, retrieved documents, emails, tickets, web pages or tool descriptions can rewrite what the model has been told to do.
  • The data the model can reach. Retrieval indexes, uploaded files, conversation memory, tenant boundaries, and what leaks back out through an answer.
  • The actions the model can take. Function calls, MCP servers, code execution, outbound HTTP, email and ticket writes, and the identity each of them runs as.
  • The way output is handled. Markdown, HTML and image rendering, link and URL construction, and downstream systems that trust model output as data.
  • The surrounding platform. API authentication, vector database access control, secret handling, rate limits, and the logs an incident would be reconstructed from.
  • The governance record. Whether the evidence produced can answer Articles 15, 26 and 50 of the AI Act, and whether the logs meet the six-month deployer floor.

Not in scope

  • A model benchmark. We do not rank base models, measure reasoning quality or score accuracy on a public dataset.
  • A bias or fairness audit. Discrimination testing is a separate discipline with separate expertise; an LLM security audit does not substitute for it.
  • A brand-visibility check. Several products sell an "LLM audit" that measures how often a model mentions your company. That is marketing analytics, not security.
  • A certification. No one certifies LLM applications today; prEN 18286, the quality management standard being drafted for AI Act purposes, is still in development.
  • A one-off guarantee. Prompt injection has no patch. An audit measures exposure at a point in time and verifies the controls that hold when the model is fooled.

The distinction matters for procurement. If a proposal promises to "make the model safe" rather than to test what the application does when the model is manipulated, it is describing a product, not an audit.

§ 02 Applicability

Three deployments that need this, and why

The trigger is not the size of the model. It is the combination of data the assistant can reach and actions it can take on behalf of someone who did not authorize them.

  1. PROFILE A A customer-facing chatbot Anything a member of the public can type into. The attacker controls the input directly, the assistant usually holds a system prompt with business logic, and it often has a retrieval index of internal documentation behind it. Since 2 August 2026, providers of systems that interact directly with people must also make clear that the counterpart is an AI system. Risk focus: LLM01, LLM02, LLM07AI Act: Article 50(1)
  2. PROFILE B An internal copilot over confidential data A RAG assistant across a document store, a mailbox, a ticket system or a code repository. The attacker never talks to the model: they place instructions in a document the model will later read. This is the shape of EchoLeak and of the Slack AI retrieval attack, and it is the hardest class to detect from logs. Risk focus: LLM01, LLM08, LLM02ATLAS: AML.T0051.001, AML.T0070
  3. PROFILE C Agents with tools and MCP servers Anything that can act: function calling, Model Context Protocol servers, coding agents, browser agents. Here a successful injection is not a disclosure, it is an unauthorized transaction. Third-party MCP servers add an unreviewed supply chain to an already trusted execution path. Risk focus: LLM06, ASI01 to ASI05ATLAS: AML.T0053, AML.T0086, AML.T0110

§ 03 Coverage

What an audit actually tests

Eight control areas, each tied to a published identifier so the report can be checked against something other than our own opinion. Every finding is filed against one of them.

  1. LLM01 Prompt injection, direct and indirect Instructions are seeded into every channel the model ingests: the chat box, uploaded files, retrieved documents, HTML fetched from the web, ticket bodies, commit messages, and the descriptions of the tools themselves. Each attempt is repeated across runs, because temperature alone changes whether a guardrail fires. MITRE ATLAS: AML.T0051, AML.T0051.001, AML.T0093, AML.T0094Evidence: transcript with the exact payload and the model response
  2. LLM02 · LLM07 Sensitive information and system prompt disclosure What the assistant will reveal about its own configuration, its retrieval sources, its tool inventory and other users. System prompt extraction matters less for the words than for what those words disclose: internal endpoints, allow-lists, business rules and, too often, credentials. MITRE ATLAS: AML.T0056, AML.T0057Evidence: recovered configuration, mapped to its source of truth
  3. LLM08 · LLM04 Retrieval poisoning and vector store weaknesses Whether a document a low-privileged user can write reaches the index, whether embeddings from separate tenants share a namespace, whether retrieval respects the same access control as the underlying store, and what the assistant answers when a poisoned passage outranks the true one. MITRE ATLAS: AML.T0070, AML.T0071, AML.T0064, AML.T0082Evidence: index write path and a retrieval trace per query
  4. LLM06 · ASI02 Excessive agency, tool and MCP abuse The complete inventory of callable actions and the identity each one runs as. We test tool invocation from untrusted content, argument tampering, chained calls, approval fatigue, and whether a tool description can itself carry instructions. MCP servers are reviewed as dependencies, with their transport and authorization handling. MITRE ATLAS: AML.T0053, AML.T0086, AML.T0110, AML.T0098Evidence: tool inventory with the privilege each call actually holds
  5. LLM03 · ASI04 Supply chain and model provenance Where the model weights, adapters, embeddings, prompt templates, agent frameworks and MCP servers come from, how they are pinned, and who can change them without review. An AI bill of materials is the deliverable most teams find they cannot produce. Reference: CVE-2025-6514 in mcp-remoteEvidence: component inventory with source, version and update path
  6. LLM05 Output handling and rendering Model output is untrusted input to whatever consumes it. We test markdown and HTML rendering, image and link construction, and any downstream parser that treats a response as structured data. This is the channel that turns a disclosure into an exfiltration. MITRE ATLAS: AML.T0067, AML.T0077Evidence: rendered payload and the network request it produced
  7. LLM10 Unbounded consumption and abuse Cost and availability, treated as security properties: per-user and per-tenant limits, context inflation, recursive agent loops, and whether an unauthenticated visitor can drive inference spend. Agent loops make this materially worse than a classic rate-limit test. MITRE ATLAS: AML.T0034.002Evidence: observed limits and the point at which they stop applying
  8. ART. 15 · 26 Governance and AI Act evidence The audit closes by mapping what was tested onto the obligations that apply. Article 15 requires high-risk systems to be resilient against attempts to alter their use, outputs or performance, and names data poisoning, model poisoning, adversarial examples and confidentiality attacks explicitly. Article 26 requires deployers to keep automatically generated logs for at least six months. Reference: Regulation (EU) 2024/1689, Articles 12, 15, 26, 50Evidence: control-to-article matrix and a logging gap list

The OWASP Top 10 for LLM Applications 2025 and the OWASP Top 10 for Agentic Applications 2026 are published by the OWASP GenAI Security Project; MITRE ATLAS identifiers come from the public ATLAS data set. Both are cited so the coverage claim is auditable.

§ 04 Evidence log

What has actually gone wrong, with dates

None of these were model failures in the abstract. Each was an application that let untrusted content reach a model that held private data and could reach the outside world.

  1. GitHub MCP toxic agent flow

    An instruction placed in a public GitHub issue was read by an assistant connected to the GitHub MCP server, which then copied private repository content into a public pull request. The MCP server was not vulnerable; the agent architecture was.

    Indirect injection · agent tool invocation Source Invariant Labs

  2. EchoLeak, Microsoft 365 Copilot

    A crafted email, never opened by the victim, was processed by Copilot retrieval and caused it to disclose data over the network. Microsoft scored the AI command injection at 9.3; NVD scored the same record at 7.5. Fixed server-side; no exploitation in the wild was reported.

    CVE-2025-32711 · CWE-74 Source The Hacker News

  3. Supabase MCP via a support ticket

    A support ticket carrying instructions was read by an assistant whose database MCP connection ran with the service role, bypassing row-level security. It read an integration token table and wrote the secrets back into the ticket thread.

    service_role · RLS bypass Source General Analysis

  4. mcp-remote command injection

    Connecting a client to an untrusted MCP server allowed operating-system command injection through a crafted authorization endpoint URL in the server response. A dependency in the agent supply chain, not a prompt problem.

    CVE-2025-6514 · CVSS 9.6 (JFrog) Source NVD

  5. Perplexity Comet agentic browser

    Hidden text in a Reddit comment hijacked the browsing agent. The published proof of concept read the user's email address and a one-time code from webmail and posted them back as a reply. Brave recommended separating trusted from untrusted context and requiring explicit confirmation for sensitive actions.

    Indirect injection · outbound channel Source Brave

  6. ForcedLeak, Salesforce Agentforce

    A Web-to-Lead form field carried the injection. CRM records were exfiltrated as image requests to a previously allow-listed domain that had expired and was re-registered for about five dollars. Patched by enforcing trusted URLs.

    CVSS 9.4 (Noma Security) Source The Hacker News

  7. GTG-1002 AI-orchestrated intrusion campaign

    Anthropic reported a state-sponsored group driving an agent with MCP tooling through reconnaissance, exploit generation, credential harvesting and exfiltration, with the model performing an estimated 80 to 90 percent of the campaign and humans at four to six decision points. Roughly thirty targets; a small number were breached.

    Agent misuse at operator scale Source Anthropic

§ 05 Method

How the engagement runs

Five phases. Each one produces an artifact that ends up in the report, so a reader can see what was covered and, just as importantly, what was not.

  1. Architecture and data-flow intake A working session that maps every input channel, every retrieval source, every tool and MCP server, memory, and every outbound path. Most of the eventual findings are visible on this diagram before a single payload is sent. Leaves behind A data-flow diagram with trust boundaries and the tool inventory
  2. Threat model The architecture is turned into a ranked list of applicable OWASP LLM and Agentic risks with their MITRE ATLAS techniques, and into the test plan. This is the step the bench on this page reproduces in miniature. Leaves behind A test plan tied to named risk identifiers, agreed before testing
  3. Adversarial testing Manual red teaming against the full payload taxonomy in the OWASP AI Testing Guide, supported by tooling of the garak and promptfoo class for breadth. Multi-turn, multi-language, multimodal and cross-tenant cases are run repeatedly, because a single pass measures luck rather than exposure. Leaves behind Reproducible transcripts, one per finding, with run counts
  4. Platform and integration review API authentication and authorization, vector database access control, MCP server configuration and transport, secret handling, egress rules, rate limits and the logging pipeline. This is ordinary application security work, and it is where the durable fixes live. Leaves behind Configuration findings with the exact setting and the fixed value
  5. Report, retest and mapping Findings with severity and reproduction steps, a coverage matrix against OWASP and ATLAS, the AI Act evidence pack, and an executive summary that states residual risk in plain language. A retest window is included; the retest letter records what was verified fixed. Leaves behind The report, the coverage matrix, the evidence pack, the retest letter

§ 06 Deliverables

What lands on your desk

A finding that cannot be reproduced is an opinion. Everything below is written so an engineer can re-run it and a regulator can read it.

  1. D1 Findings with reproduction transcripts Each finding carries the exact input, the channel it entered through, the model response, the run count, and the impact demonstrated. Severity reflects what the assistant could reach and do, not the novelty of the payload.
  2. D2 OWASP coverage matrix Every LLM01 to LLM10 and, for agentic systems, every ASI01 to ASI10 entry marked tested, not applicable or not tested, with the reason. The blanks are the part that makes the matrix useful.
  3. D3 MITRE ATLAS technique mapping Findings mapped to ATLAS technique identifiers so that detection engineering and threat intelligence can consume them alongside ATT&CK.
  4. D4 AI Act evidence pack A record of the robustness and cybersecurity testing performed against Article 15, a logging check against Articles 12 and 26 including the six-month retention floor, an Article 50 disclosure check, and input for the Article 6(4) classification memo.
  5. D5 Executive summary Two pages that a board can read: what was tested, what was found, what the residual risk is, and what changes when the recommendations are implemented. No severity inflation, no scores invented to fill a chart.
  6. D6 Retest and letter A retest of the fixed findings within the agreed window, and a letter stating what was verified. Useful for customers and insurers who ask for one, and honest about anything still open.

§ 07 Scope drivers

What decides the effort

There is no useful list price for this work, and any vendor publishing one is quoting for a scope they have not seen. These are the variables that actually move the number.

  1. S1 Number of applications and models One chatbot on one model is a different engagement from six assistants sharing a gateway, three model providers and a self-hosted fallback.
  2. S2 Feature surface Retrieval, tools, memory, browsing, code execution and multi-agent orchestration each add a class of tests. The bench below lists which ones your selection triggers.
  3. S3 Access level Black box takes longer and finds less. Grey or white box with the system prompt, the tool schemas and a staging environment produces materially better coverage for the same days.
  4. S4 Environment and tenancy Testing in production constrains destructive cases and needs a written rules-of-engagement annex. Multi-tenant deployments add cross-tenant isolation testing that single-tenant ones do not need.
  5. S5 Compliance driver An AI Act evidence pack, a DORA threat-led exercise or a customer security questionnaire each change the documentation burden, sometimes more than the testing itself.
  6. S6 Retest and cadence A single assessment is a snapshot. Model version changes, prompt changes, new retrieval sources and new tools each invalidate part of it, which is why retesting is scoped up front.

As a rough shape rather than a quotation: a single chatbot is a matter of days, an agent and MCP estate a matter of weeks. Drafting a scope takes about ten minutes and produces a brief you can send to any provider, not only to OffSeq.

§ 08 Interactive

The bench: threat model and obligation finder

Two working tools. The first turns an architecture into the risks and tests that apply to it. The second turns a role and a use case into the AI Act articles and dates that bind you. Nothing you select leaves your browser.

Interactive mode is not available. You can read the full reference content below. No answers are assessed and no result is calculated.

Both tools need JavaScript. The reference material behind them is below in full: the risk catalog the threat model draws from, and the regulatory dates the obligation finder uses.

OWASP Top 10 for LLM Applications 2025, the catalog every finding in an audit is filed against:

  1. LLM01:2025 Prompt Injection – user or retrieved content rewrites the model's instructions.
  2. LLM02:2025 Sensitive Information Disclosure – the assistant reveals data it should not.
  3. LLM03:2025 Supply Chain – models, adapters, frameworks and MCP servers of unverified provenance.
  4. LLM04:2025 Data and Model Poisoning – training, fine-tuning or retrieval data manipulated by an attacker.
  5. LLM05:2025 Improper Output Handling – model output consumed as trusted input downstream.
  6. LLM06:2025 Excessive Agency – more tools, permissions or autonomy than the task needs.
  7. LLM07:2025 System Prompt Leakage – configuration, rules and secrets recovered from the prompt.
  8. LLM08:2025 Vector and Embedding Weaknesses – retrieval poisoning and cross-tenant leakage.
  9. LLM09:2025 Misinformation – confident output relied on without verification.
  10. LLM10:2025 Unbounded Consumption – cost and availability abuse, including agent loops.

EU AI Act dates as they stand after Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026:

  • 2 February 2025 – prohibited practices and the AI literacy duty apply.
  • 2 August 2025 – general-purpose AI model obligations and the governance rules apply.
  • 2 August 2026 – general application, including the Article 50 transparency duties.
  • 2 December 2026 – the four-month marking grace for systems placed on the market before 2 August 2026 ends, and two further prohibited practices apply.
  • 2 December 2027 – high-risk obligations for Annex III systems apply, moved from 2 August 2026.
  • 2 August 2028 – high-risk obligations for Annex I product-embedded systems apply.
Share this threat model
Feature surface

Select what your application can actually do today. The default selection is the most common European deployment: an assistant over internal documents that can call a couple of tools.

Threat model

The three ingredients present in every published agent incident:

  • Present Not indicated Private data in reachSomething worth stealing is inside the context window
  • Present Not indicated Untrusted content in contextContent an attacker can write reaches the model
  • Present Not indicated Outbound channelThe model can move data or act outside the chat

Full path

A complete exfiltration path exists

All three ingredients are present at once, which is the exact shape of EchoLeak, the Supabase MCP case and ForcedLeak. Testing should start with an end-to-end exfiltration attempt, not with individual payloads.

Partial path

Two of three ingredients are present

One link is missing today. That link is usually added by the next feature request, so the architectural controls are worth putting in before it is, rather than after.

Single

One ingredient is present

Exposure is currently limited to what a manipulated answer can do on its own: misinformation, disclosure of the prompt, and abuse of inference cost. Still worth testing, at a smaller scope.

Model only

No architectural path indicated

On this selection the assistant is a text interface with no private data, no attacker-writable content and no way to act. The remaining risks are the ones that apply to every deployment: prompt disclosure, output handling and consumption.

13 risk areas apply to this architecture

Share this threat model

Opens the OffSeq scoping brief. Your selections stay in this page; copy the applicable identifiers into the brief if you want them included.

  1. LLM01:2025 Direct prompt injection Audit test The full payload taxonomy from OWASP AITG-APP-01: role play, context hijacking, obfuscation and token smuggling, multi-language, payload splitting, structured-output attacks and multi-turn escalation, each repeated across runs. ATLAS AML.T0051, AML.T0065, AML.T0068 Precedent Universal to every deployment; OWASP calls it the joint that connects the other failures
  2. LLM01:2025 Indirect injection through ingested content Audit test Instructions are planted in every channel the model reads without a human deciding to read it: a document in the index, a page it will fetch, an uploaded file, a ticket body, an MCP tool description. Delayed-trigger payloads are included. ATLAS AML.T0051.001, AML.T0093, AML.T0094 Precedent EchoLeak (CVE-2025-32711), Perplexity Comet, ForcedLeak
  3. LLM07:2025 System prompt and configuration leakage Audit test Direct and oblique extraction of the system prompt, tool schemas, retrieval configuration and guardrail wording, followed by an assessment of what the recovered text discloses: endpoints, allow-lists, business rules, credentials. ATLAS AML.T0056
  4. LLM02:2025 Sensitive information disclosure Audit test Whether the assistant will surface documents, fields or other users' data that the requesting identity has no right to, including through summarization, translation and quotation rather than direct retrieval. ATLAS AML.T0057 Precedent Slack AI retrieval disclosure, recorded as an ATLAS case study
  5. LLM08:2025 Retrieval poisoning and vector store weaknesses Audit test Who can write into the index, whether retrieval enforces the same access control as the source store, whether a planted passage can outrank the authoritative one, and whether embeddings from different tenants share a namespace. ATLAS AML.T0070, AML.T0071, AML.T0064, AML.T0082 Precedent Morris II research worm, recorded as an ATLAS case study
  6. LLM04:2025 Data poisoning through user-supplied content Audit test Whether content submitted by a low-privileged user reaches a fine-tuning set, a feedback loop, a shared memory or an index that higher-privileged users query, and what review stands between submission and ingestion. ATLAS AML.T0070, AML.T0080
  7. LLM05:2025 Output handling and rendering exfiltration Audit test Markdown and HTML rendering, image and link construction, and any downstream consumer that parses the response. The question is whether a manipulated answer can cause a request to leave the browser or the server. ATLAS AML.T0067, AML.T0077 Precedent ForcedLeak exfiltrated CRM records as image requests
  8. LLM06:2025 · ASI02:2026 Excessive agency and tool misuse Audit test A complete inventory of callable actions, the identity each runs as, and whether untrusted content can trigger one. Argument tampering, chained calls and write actions with no human confirmation are tested individually. ATLAS AML.T0053, AML.T0086 Precedent GitHub MCP toxic agent flow; Supabase MCP with the service role
  9. ASI03:2026 Identity and privilege abuse across tools Audit test Whether each tool call carries the end user identity or a shared service identity, how tokens are stored and scoped, and what a single compromised tool call can reach in the systems behind it. ATLAS AML.T0098, AML.T0082 Precedent Supabase MCP running with a row-level-security bypass role
  10. ASI04:2026 · LLM03:2025 Agentic supply chain and model provenance Audit test Provenance and pinning for models, adapters, embeddings, prompt templates, agent frameworks and MCP servers; who can change each without review; and whether an AI bill of materials can be produced at all. ATLAS AML.T0104, AML.T0110 Precedent mcp-remote command injection, CVE-2025-6514
  11. ASI09:2026 Human-agent trust exploitation Audit test How confirmations are presented, whether the approval dialog shows the real action and its real arguments, and how many approvals a normal working day produces before the human stops reading them. ATLAS AML.T0067
  12. LLM10:2025 Unbounded consumption Audit test Per-user, per-tenant and global limits; context inflation; recursive agent loops; and whether an unauthenticated visitor can drive inference spend or exhaust a shared quota. ATLAS AML.T0034.002
  13. LLM09:2025 Misinformation and unsafe reliance Audit test Whether answers are grounded in retrievable sources, whether citations resolve, and what the interface does when the model has no basis for an answer. Assessed as a security property where the output drives a decision or a transaction. ATLAS Not an adversarial technique; a reliance failure

§ 09 Calendar

EU AI Act dates, after the Digital Omnibus

Most published guidance still says high-risk obligations began on 2 August 2026. That stopped being true on 27 July 2026. These are the dates as amended, each taken from the Official Journal text.

  1. Regulation (EU) 2024/1689 enters into force

    The AI Act was adopted on 13 June 2024 and published in the Official Journal on 12 July 2024.

    In force

  2. Prohibited practices and AI literacy

    Chapters I and II apply: the Article 5 prohibitions and the Article 4 AI literacy duty.

    In force

  3. General-purpose AI models and governance

    Obligations for providers of general-purpose AI models and the governance and penalty framework begin to apply. The General-Purpose AI Code of Practice was published on 10 July 2025.

    In force

  4. Digital Omnibus on AI enters into force

    Regulation (EU) 2026/1744 of 8 July 2026, published in the Official Journal on 24 July 2026, enters into force on the third day after publication and amends the application dates below.

    In force

  5. General application and Article 50 transparency

    The general date of application, unchanged by the Omnibus. Chatbots must tell people they are talking to an AI system, and generated content must be marked and labeled. The Commission put the enforcement ceiling for these rules at EUR 15 million or 3 percent of worldwide turnover.

    In force

  6. Marking grace ends; two further prohibitions apply

    Providers of generative systems placed on the market before 2 August 2026 must comply with the Article 50(2) marking obligation by this date. Two prohibited practices added by the Omnibus also begin to apply.

    Pending

  7. Annex III high-risk obligations

    Chapter III Sections 1 to 3 apply to systems classified as high-risk under Article 6(2) and Annex III, including recruitment, workforce management, creditworthiness and insurance pricing. Moved from 2 August 2026.

    Pending

  8. Annex I product-embedded high-risk obligations

    The same obligations apply to systems classified as high-risk under Article 6(1) and Annex I, the safety components of regulated products. Moved from 2 August 2027.

    Pending

Every date above is taken from the consolidated Official Journal texts of Regulation (EU) 2024/1689 and Regulation (EU) 2026/1744, cross-checked against the European Commission's implementation page. Where a secondary source disagrees, the Official Journal wins.

§ 11 Questions

Questions we are asked before every engagement

Short answers. The longer versions are in the annexes.

What is an LLM security audit?
A structured adversarial and configuration review of an LLM-based application: its prompts, retrieval sources, tools, outputs, hosting and governance, measured against the OWASP Top 10 for LLM Applications 2025, the OWASP Top 10 for Agentic Applications 2026 and MITRE ATLAS. It is not a model benchmark, a bias audit, or the brand-visibility product also sold under the name "LLM audit".
How is it different from an ordinary penetration test?
A web penetration test asks whether input can be injected into SQL or a shell. An LLM audit asks whether input can be injected into the model's instructions through a document, an email, a web page or a tool description, and then asks what the model is allowed to do once it has been. Ordinary API, authentication and infrastructure testing stays in scope, because that is where most of the fixable findings are.
Do we need this if we only use ChatGPT or Copilot at work?
You are a deployer, so the Article 4 AI literacy duty has applied to you since 2 February 2025. An audit becomes warranted at the point where the assistant can reach confidential data or take actions: EchoLeak and the Supabase MCP case were both ordinary corporate deployments of commercial products.
When does the EU AI Act apply to us?
If you place an AI system on the EU market or use one under your authority in the EU, regardless of where the company is established. What you must do depends on your role, provider or deployer, and on the risk class of the system. The obligation finder on this page walks through both.
Did the high-risk deadline really move?
Yes. Regulation (EU) 2026/1744, published in the Official Journal on 24 July 2026, moved the Annex III high-risk obligations to 2 December 2027 and the Annex I product-embedded ones to 2 August 2028. Article 50 transparency was not moved: it has applied since 2 August 2026, with a four-month marking grace to 2 December 2026 for systems already on the market.
Is our HR or credit assistant high-risk?
Annex III area 4 covers recruitment, filtering of applications, promotion and performance monitoring; area 5 covers creditworthiness and life and health insurance pricing. Article 6(3) allows a narrow derogation for procedural or preparatory tasks, but a system that performs profiling of natural persons is always high-risk. Either way, a provider that concludes the system is not high-risk must document that assessment.
What is prompt injection, and can it be fixed?
The model receives instructions and data in the same token stream and has no reliable way to tell them apart, so there is no patch in the sense of a version that removes the class. An audit measures how much exposure the architecture creates and verifies the controls that hold when the model is fooled: least privilege on tools, allow-listed actions, human confirmation for anything with consequences, output encoding and egress control.
Does ISO/IEC 42001 make us AI Act compliant?
No. ISO/IEC 42001 is a voluntary, certifiable management-system standard for AI; the AI Act is binding law with different obligations. The European standard being drafted for AI Act purposes, prEN 18286, is intended to map to the ISO/IEC 42001 Annex A controls and, once cited in the Official Journal, to give a presumption of conformity for the Article 17 quality management requirement. It is not published yet.
What do we actually receive?
Findings with reproducible transcripts and severity, an OWASP coverage matrix that also records what was not tested, a MITRE ATLAS technique mapping, an AI Act evidence pack for Articles 12, 15, 26 and 50, a two-page executive summary and a retest letter.
How long does it take, and how often should we repeat it?
A single chatbot is typically a matter of days; an agent and MCP estate a matter of weeks. Retest after a model version change, a system prompt change, a new retrieval source or a new tool, since each of those invalidates part of the previous result. Annually is the floor, and ahead of a high-risk application date if one applies to you.

Sources

  1. Regulation (EU) 2024/1689 (Artificial Intelligence Act) EUR-Lex, Official Journal of the European Union · 2024 Articles 3, 4, 6, 12, 15, 25, 26, 50, 99 and Annex III as cited on this page.
  2. Regulation (EU) 2026/1744 (Digital Omnibus on AI) EUR-Lex, Official Journal of the European Union · 2026 Amended application dates: 2 December 2027 for Annex III, 2 August 2028 for Annex I, and the four-month Article 50(2) marking grace.
  3. AI Act implementation timeline and scope European Commission, Directorate-General for Communications Networks, Content and Technology · 2026
  4. Transparency obligations under Article 50: frequently asked questions European Commission · 2026
  5. Safer and more transparent AI European Commission · 2026
  6. OWASP Top 10 for LLM Applications 2025 OWASP GenAI Security Project · 2025
  7. OWASP Top 10 for Agentic Applications 2026 OWASP GenAI Security Project · 2026
  8. OWASP AI Testing Guide, AITG-APP-01: Testing for Prompt Injection OWASP · 2025
  9. MITRE ATLAS technique and case-study data MITRE · 2026 Source of the AML technique identifiers and the Slack AI and Morris II case studies cited above.
  10. Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile (NIST AI 600-1) National Institute of Standards and Technology · 2024