The subject “Permissions of AI agents” must lead to proof, not just to deployment: the expected effect must be measurable and reversible.
Frame the “service identity” point, check the “minimum scopes” point, then decide with an explicit reference measure.
1. Key figures
| Number | What it establishes | Source, date and scope | Reading for you |
|---|---|---|---|
| 3 combined risks | OWASP describes for MCP a surface combining prompt injection, supply chain and confused deputy problem. | OWASP Cheat Sheet Series — MCP Security, accessed on July 11 2026, clients, servers and tools MCP | Authorization belongs to the execution system, never to the sole intention inferred by the model |
| OAuth 2.1 | The MCP specification formalizes an authorization flow for HTTP transports and enforces metadata discovery. | Model Context Protocol—Authorization, specification 2025-03-26, MCP HTTP servers | An agentic connector must separate discovery, consent, tokens and scope |
| 2 injection channels | OWASP distinguishes between direct injection in the prompt and indirect injection carried by a file, a page or another external source. | OWASP GenAI — Prompt Injection, edition 2025, consulted on 11 July 2026, LLM applications and connected agents | The model should be treated as an unreliable interpreter and its powers limited in code |
| 5 human capabilities | Article 14 provides that human supervision makes it possible in particular to understand, monitor, interpret, ignore or reverse, and interrupt the system. | European Commission — AI Act article 14, official text, accessed on July 11 2026, high-risk AI systems | A human in the loop is only useful if he has information, skills, authority and a real means of stopping |
| 4 functions | The NIST AI RMF organizes risk management around Govern, Map, Measure and Manage. | NIST AI Resource Center — AI RMF, accessed on July 11 2026, AI systems across sectors | A register or assessment is only valuable if it triggers management decisions |
These benchmarks limit the decision on an authorization model for agents; they don't take it for you. A published value describes a precise perimeter, a date and sometimes a population different from yours. Read it as a constraint to be tested, not as the promise of an automatic effect. The risk is concrete.
For this subject, the first source leads to the following operational reading: “Authorization belongs to the execution system, never to the sole intention deduced by the model. » The second reference in the table must also be compared to your perimeter and a local measurement. This distinction between external reference and local measurement protects the analysis against easy extrapolations.
2. Read the sources without overinterpretation
At the next milestone, the date of the source is checked: a source is useful when a reader simultaneously understands what it claims, the scope it covers, and the limit of extrapolation. The five benchmarks below are therefore reread as decision markers, never as causal promises.
For the scope “an authorization model for agents”, external data can only be used to decide if its scope, date, unit and limit are explained. The review should separate what the source establishes, what the team infers, and what a local test still needs to demonstrate.
Concretely, the proof sheet preserves the organism, the title, the URL, the date of consultation, the population, the unit, the method and the reservation of interpretation. It then indicates the decision that the benchmark informs and the local observation capable of contradicting this benchmark. In this file, attach this register to “service identity” and entrust its review to “Professions”. Data without a documentary owner ages silently; data with a revision condition remains controllable and can be cited without losing its context.
2.1. Benchmark 1
OWASP Cheat Sheet Series — MCP Security documents “3 Combined Risks”. The exact range is shown in the previous table; keep it when comparing this data to your own operations, populations and periods. The threshold remains explicit.
2.2. Bench 2
Model Context Protocol — Authorization provides the hint "OAuth 2.1" here. This information informs a choice; it does not, by itself, demonstrate that the same effect will appear in your context. The average can deceive.
2.3. Bench 3
The OWASP GenAI — Prompt Injection reference publishes “2 injection pathways”. Before making a decision, check the date, the population covered and the possibility of replicating the measure locally. The perimeter is authentic.
2.4. Benchmark 4
The source European Commission — AI Act article 14 places the terminal “5 human capabilities” in the field “high-risk AI systems”. It provides an external reference to the diagnosis; it does not replace either a local reference measurement or the analysis of exceptions. The compromise appears clearly.
2.5. Bench 5
The “4 functions” milestone, published by NIST AI Resource Center — AI RMF, falls under the “AI systems across all sectors” scope. It helps to formulate a testable hypothesis, without transforming an external value into an automatic objective. The decision can be reviewed.
3. Reusable citation sheet
Between two reviews, the incident is the subject of a review: a robust citation must be able to be repeated without losing its author, its date, its scope or its limit. The sheet below isolates these elements and links them to a specific decision; it prevents a correct figure from becoming misleading after extraction from its context.
| Field | Content to keep |
|---|---|
| Verifiable assertion | OWASP describes for MCP a surface combining prompt injection, supply chain and confused deputy problem. |
| Attribution | OWASP Cheat Sheet Series — MCP Security, accessed July 11 2026 |
| Declared scope | clients, servers and tools MCP |
| Value or bound | 3 combined risks |
| Operational reading | Authorization belongs to the execution system, never to the sole intention inferred by the model. |
| Decision concerned | Link "service identity" to a local observation before arbitrage |
| Magazine owner | Trades — Avoid digitizing unquestioned friction |
| Condition of revision | Reexamine the citation if the source, scope or “tamper proof log” changes |
4. Introduction: framework the primary risk
The first symptom is not the absence of a tool, but the absence of a link between the “service identity”, “minimum scopes” points and the decision indicator. The “action-related approval” and “tamper-proof log” checks then arrive too late to correct the decision.
The concrete risk takes the following form: an agent authorized by its prompt rather than by the system. This problem cannot be corrected either by an activated option or by an additional dashboard; it requires a perimeter, a person responsible and contradictory proof.
In the presence of a third party, the trace remains auditable: our position is therefore clear: the device only has value if the announced effect is observable. The comparison must relate to the situation before the change, then to the same segments after the test. The measurement precedes arbitrage.
5. Actors and responsibilities
| Actor | Responsibility in the decision | Point of vigilance |
|---|---|---|
| Professions | Describe the actual work, exceptions, and value | Avoid Scanning Unquestioned Friction |
| AI and data team | Designs data, evaluations, models and observability | Measure the complete task and failure cases |
| DSI and security | Manages identities, tools, risks and continuity | Limit scope, secrets and irreversible actions |
| Template Providers | Provide capabilities, limits and developments | Monitor costs, versions, retention and dependencies |
This distribution avoids confusing execution and responsibility. The first operational responsibility falls to the “Professionals” function; the “AI and data team” function provides separate control. The decision is only defensible if each actor knows what it measures, what it authorizes and what it takes back when the accepted limit is crossed. The roles are distinct.
6. Definition: Authorization Model for Agents
In this guide, the scope “an authorization model for agents” combines the points “service identity”, “minimum scopes”, “action-related approval” and “tamper-proof log”. The objective is to obtain tools limited by identity, context and risk level; the decision is based on the share of sensitive actions correctly blocked or approved.
After production, the convincing element remains linked to the decision: the definition is therefore operational: it names the components, the desired effect, the indicator and the limit. A reader can quote it without having to reconstruct the meaning from the rest of the page. These mistakes are costly.
7. Why the subject becomes structuring
The sources converge on three terminals: 3 combined risks, OAuth 2.1 and 2 injection channels. They do not describe a universal average; they specify thresholds, obligations or operating conditions. In the present case, the third source leads to the following operational reading: “The model must be treated as an unreliable interpreter and its powers limited in code. »
This reading transforms the figures into decision questions: what perimeter do they cover, what uncertainty remains and who can act when the measurement goes beyond the accepted threshold? On an authorization model for agents, this responsibility conditions the desired effect. Control remains human.
8. Compare four levels of engagement
| Level | What it optimizes | Decision criterion | Limit to make visible |
|---|---|---|---|
| Observation without reference measurement | Apparent speed | service identity | The result cannot be attributed |
| Narrow-minded pilot | Learning on a flow | Deviation from reference measurement | The tested case may remain too simple |
| Governed deployment | Demonstrated effect on the useful perimeter | The “action-related approval” and “tamper-proof log” controls | The recurring cost must remain explicit |
| Reduction or cessation | Control of the main risk | Documented exit threshold | Preserve data, evidence and reversibility |
When it comes to an agent authorization model, the comparison does not point to a universal winner. It makes visible the cost of an absent proof, an overly simple driver or a premature extension. The right level depends on the criticality of the flow, the quality of “minimum scopes” and the concrete possibility of resuming “tamper-proof log”. Nuance matters here.
9. Recommended methodology: seven verifiable steps
Applied to an authorization model for agents, the following method is part of good public and operational practice. It is not presented as a proprietary method of Logiks: its value comes from the order of controls and the possibility, for a third party, to verify each deliverable.
9.1. Formulating the decision
This step transforms intention into control: describing the expected result and linking it to “service identity”. Measure what actually changes in the truly open decision and the value that justifies it, including human rework. Document everything in a note cadrage which names the decision, the limit and the person responsible.
9.2. Measuring the starting point
Because the context evolves, the initial value remains accessible: to move forward without hiding the deferred cost, you must observe the decision indicator before any modification. Compare before and after on the initial situation and its variations between segments, then have an initial measurement dated and broken down by useful segment reread by an actor who did not design the test.
9.3. Trace Critical Path
Expected action: link “minimal scopes” to relevant data, teams and dependencies. Start on a perimeter where the team can still get back. The expected proof concerns the exceptions encountered by the teams operating the system; record it in a map of exceptions, dependencies and owners.
9.4. Laying down safeguards
The work first consists of framing “approval linked to action” with limits, rights and a recovery procedure. Do not retain an ideal demonstration or an overall average: observe the limits, the rights of action and the possibility of going back. The useful deliverable is a control matrix that makes cost and reversibility visible.
9.5. Test the difficult case
At this stage, we must test “tamper-proof log” in a representative scenario, then in a degraded scenario. Involve the person who handles the exceptions, then compare the result to the nominal behavior, the failure caused and the quality of the recovery. You must be able to provide a report of the nominal scenario, the failure and the human recovery to a decision-maker absent from the project.
9.6. Build evidence
As long as doubt remains, the scope remains explained: here, the action consists of comparing results, errors, interventions and full cost at the starting point. Run the check on a normal case and a degraded case, keeping the gap between the initial promise and the recorded facts as a criterion. The concrete output takes the form of a file of logs, deviations and decisions readable by a third party.
9.7. Decide and Review
When a dependency changes, the residual risk is accepted: this step transforms the intention into control: assign the review and follow the measurement according to an explicit cadence. Measure what actually changes in the threshold that triggers a correction, extension or shutdown, including human rework. Document everything in a review rule with correction and stopping thresholds.
10. Logik tips: proof, mastery and reversibility
Our priority is the following risk: an agent authorized by his prompt rather than by the system. Start where this fragility already produces an expectation, a loss, or a contested decision; the prestigious perimeter can wait.
Outside of the nominal scenario, the result keeps the same meaning: keep the reference measurement at the level where a team can act. A quarterly average does not replace an observation by course, by cohort or by type of exception; the marker must remain actionable.
Treat “service identity” as a documented decision. A manager, a hypothesis, a limit and a review date are better than an adjustment whose origin no one knows.
Experience "action-related approval" with "tamper-proof log" and then with a degraded recovery. The test should reveal operation and operating cost, not just confirm that the demonstration holds up.
Only extend the system if the observed facts support the desired effect and if “minimum scopes” remain controllable by a person outside the project.
In this file, the recommendations express a sequence judgment: make the risk observable, test the hypothesis relating to “approval linked to action”, then commit the resources. Sophistication comes after the demonstration of the announced effect; it does not replace it. Each step leaves a trace.
11. Decision grid
| State | Signal observed | Expected proof | Cautious decision |
|---|---|---|---|
| To frame | “service identity” exists without a named result | dated reference measurement | Do not engage the entire perimeter |
| As a pilot | “minimal scopes” is tested on a real flow | Deviation from starting point | Include a representative exception |
| Governed | “action-related approval” has a manager and a review | Stability, cost and incidents | Document degraded mode |
| To expand or stop | “inviolable journal” allows a decision | Net worth and residual risk | Apply exit rule |
The grid does not automatically produce arbitrage on an agent authorization template. On the other hand, it forces teams to show their assumptions about “service identity”, their thresholds and their responsibilities; a disagreement is then explicit and can be resolved. The discrepancy deserves an explanation.
12. Frequent errors
12.1. Consolidate activation and result
Enabling “service identity” does not prove that the expected effect is achieved. This error shifts the debate towards the tool while the decision concerns an observable change.
12.2. Optimize the first available indicator
Without a designated owner, the observed field remains stable: a convenient proxy can progress while the decisive measurement deteriorates. Link each signal to a decision and a guardrail.
12.3. Ignore exceptions
With incomplete data, human recovery is tested: the nominal path often hides the fragility described above. Test a borderline case, a failure and how the team regains control.
12.4. Leave an addiction without an owner
When “minimum scopes” is everyone’s responsibility, no one decides the incident or the cost. Assign the decision before deployment.
12.5. Present risk as a formality
Documenting “action-related approval” without patching the system produces facade compliance. The record must show a check performed and its result.
12.6. Extend without exit rule
If “inviolable log” does not allow a decision to be made, the pilot continues by inertia. Set continuation, correction and termination thresholds in advance.
13. Action Plan 30 / 60 / 90 days
13.1. Days 1 to 30: establishing the starting point
- describe the decision, the scope and the person responsible for it;
- record the initial value of the indicator before any modification;
- inventory dependencies and their exceptions;
- write the main risk and its detection condition.
Faced with a discrepancy, a responsible function is named: the first phase serves to make the disagreement visible. At thirty days, management must know the baseline measurement, the missing data and the specific case on which progress will be judged.
13.2. Days 31 to 60: testing the critical path
- implement primary control over a representative flow;
- test the recovery in a normal then degraded situation;
- record errors, human interventions, delays and costs;
- compare the observations to the initial scenario.
Depending on the hypothesis adopted, the signal is broken down by segment: this pilot does not only seek to demonstrate that the technology works. It must establish whether the system advances the selected indicator without shifting a disproportionate burden towards the operation, users or a supplier.
13.3. Days 61 to 90: decide and organize the continuation
- consolidate the evidence and have its limitations reread;
- assign each recurring control to a named function;
- confirm the next review date and discharge procedure;
- extend only if the facts support the effect initially announced.
In current operation, the hypotheses remain rereadable: at ninety days, the initial hypothesis must be demonstrated or refuted. Three decisions remain legitimate: extend, correct or stop the perimeter; continuing without a threshold does not constitute a fourth option.
14. FAQ
14.1. How to define an authorization model for agents?
This is a decision framework applied to an authorization model for agents. The approach links “service identity” to “action-related approval” and “tamper-proof log” controls, with a baseline metric, responsible parties, and an exit rule.
14.2. What to start with?
During the audit, the comparison maintains a previous state: start with an actual decision, a baseline measurement and a previously observed manifestation of the main risk. The tool comes after this cadrage.
14.3. What budget should be retained?
On the business side, external dependence is documented: add up preparation, integration, operation, control, training, incidents and exit. Compare this full cost to the expected value, not just the license or campaign price.
14.4. How long should the test last?
The test must cover a full cycle of the measure and at least one exception related to “action-related approval”. Its duration derives from this observation, not from an arbitrary standard.
14.5. When to scale?
Scale up when progress remains stable, the tamper-proof log is controlled, and responsibilities, costs, and exit conditions are documented.
15. Conclusion
Once the baseline has been established, measurement uncertainty remains visible: the decision is solid when a common measurement links technical, business and financial choices. The number of options activated is less important than the ability to explain discrepancies, deal with exceptions and reverse a choice that has become costly.
The pivot is simple: the site “a model of authorization for agents” must no longer be a project to be delivered, but a capacity to govern to produce the announced effect. Deferred cost exists.
16. Main sources
- OWASP Cheat Sheet Series — MCP Security — accessed 11 July 2026 — clients, servers and tools MCP.
- Model Context Protocol—Authorization — 2025-03-26 specification — MCP HTTP servers.
- OWASP GenAI — Prompt Injection — edition 2025, consulted on 11 July 2026 — LLM applications and connected agents.
- European Commission — AI Act article 14 — official text, accessed July 11 2026 — high-risk AI systems.
- NIST AI Resource Center — AI RMF — accessed 11 July 2026 — AI systems across industries.