Logiciel Solutions Contact Us
Success Stories Tech News Investors Contact Us
whitepaper

Agentic Systems: An Engineering Reference.

An agent raises four questions ordinary software does not: whose identity it acts under, which tools it may call, what its memory is allowed to keep, and how far one bad run reaches before something stops it. This reference answers them with twenty-two controls across seven families. Each control names what it requires, where it is enforced, and the one artefact you can point at to show it is real. The OWASP agentic risk IDs sit on every family, so a reviewer gets a published reference rather than your own house opinion.

In depth

Four Questions An Agent Raises That A Web Application Never Did.

01

Whose authority the agent acts under is the first question.

Most fleets run on one shared service principal, so every action is attributable to the platform and to nobody in particular. AGT-1 asks for a directory identity per deployed agent with a named human owner who can suspend it, the initiating principal carried through the whole run by token exchange rather than a standing agent token, and reach evaluated per call as the intersection of the agent grant and the requester entitlements. An agent should never be able to read a record the person who asked for it cannot.

In shortAn agent should never be able to read a record the p…
02

Tool-level permission is too coarse for an agent.

The same call is routine with one argument and irreversible with another, which is why AGT-2 constrains values, record counts, target accounts and destination domains rather than tool names. Connectors are pinned by digest with an owner and a registered schema, and generated code runs in an isolated executor with no credentials, no egress and an ephemeral filesystem. A registry that accepts connectors at runtime changes your allowlist without anybody cutting a release.

In shortA registry that accepts connectors at runtime change…
03

The objective is the attack surface unique to agents.

Change the goal once and everything downstream stays technically authorised, since the tool calls that follow are all permitted ones made for the wrong reason. AGT-3 fixes goal, constraints and success criteria server-side at dispatch, compares them against a hash on the run record every planning turn, and measures executed actions against the declared goal so runs that wander are halted and sampled. Hold the objective in conversational context and any retrieved page can restate it.

In shortHold the objective in conversational context and any…
04

Memory turns a single injection into a durable one.

A summary written back as first-class context becomes a standing instruction that appears in no prompt anyone reviews and outlives the conversation that introduced it. AGT-4 wants source, author, session and a type on every stored item, a promotion gate between session scratch and long-term memory, and partitions by tenant, user and purpose enforced server-side. The default for anything in session scratch should be expiry, not promotion.

In shortnot promotion
The detail

Other Agents, The Radius Of A Bad Run, And The Record Afterwards.

The first four families bound one agent. These three decide what happens when it talks to other agents, when a single wrong answer becomes a thousand actions, and when somebody has to say afterwards who was answerable for it.

Zone · 01

Inter-agent trust

Trust does not transit. An agent accepting a task from another inherits that agent's compromise, so messages are signed and verified with sender identity carried rather than asserted in the payload, hops are capped with the full chain and original principal on every one, and a registry holds every participating agent with its owner, scope and permitted peers. Anything exercising agent patterns the registry does not hold raises an alert.

Zone · 02

Blast radius

Cascading failure is the agentic outage: one wrong answer becomes a thousand actions before a person reads the dashboard. Ceilings per run on tool calls, records touched, spend and wall-clock fail closed at the limit. Agents are partitioned so one cannot exhaust another's quota, queue or credential. The kill switch revokes the identity and aborts runs already executing, and it is drilled on a live agent against a clock.

Zone · 03

Replay and attribution

A completion log says what was returned. A plan trace answers why it did that: goal, plan, every decision point, tool call and result on one run ID, with model and prompt versions pinned. The proof is a replay of one named historic run from the trace alone, with a stopwatch on it. Attribution has to survive credential rotation and staff changes, so identifiers are stored rather than the credential.

By the numbers

The figures that make it a board-level conversation.

22
agent controls in the catalogue, spread across seven families and mapped to the ASI risk IDs
10
OWASP ASI entries in the 2026 agentic top ten, every one mapped to a family here
$1M
added to the average breach where the attack was AI-driven
Inside the report

What you'll take away.

01

Step 1 - Pin the objective where the planner cannot rewrite it

Goal, constraints and success criteria fixed server-side at dispatch, re-asserted each planning turn, and hashed on the run record. A monitored drift ratio catches the runs that wander off it.

02

Step 2 - Make the stop reach work already in flight

A feature flag blocks new runs and lets dispatched ones finish holding valid credentials. Pair identity revocation at the tool gateway with an abort path the run controller honours mid-plan, then time the drill.

03

Step 3 - Require an author on every memory write

Source, author, session and a type marking the item data rather than instruction, refused at the schema if absent. Three required fields on the write path stop a retrieved instruction becoming a permanent preference.

04

Step 4 - Constrain arguments, not just tool names

Value ceilings, record counts, target accounts and destination domains written as policy the gateway evaluates per invocation, with one CI case behind each constraint. Same call, routine or irreversible depending on what is passed.

Questions

Frequently asked.

What does an agent need that an ordinary AI feature does not?

An identity of its own, argument-level limits on the tools it calls, an objective it cannot be talked out of, memory with an author on every entry, rules for talking to other agents, a bounded run, and a trace you can replay. Seven families, twenty-two controls.

Which three controls are missing most often?

An objective the agent cannot be talked out of, a stop that reaches work already in flight, and memory writes that carry an author. All three are cheap while the orchestrator is being written and expensive once it is live, since each asks it to hold state it never held.

Is a shared service account for our agent fleet a problem?

Yes, and it is the common one. One principal across a fleet makes every action attributable to the platform and to nobody, so nothing can be suspended, reviewed or explained per agent. Give each deployed agent a directory identity with a named human owner, registered by the deploy job.

Should we wait for the NIST agentic control overlays?

No. The single-agent and multi-agent overlays under COSAiS are still in draft, built on SP 800-53, so expect a second vocabulary later rather than a different set of controls. Cite the OWASP ASI identifiers now and map across when the overlays are published.

Does this cover model and application security as well?

No, deliberately. Trust zones, output handling, retrieval permissions and the model supply chain are assumed here rather than repeated, and they sit in our AI security reference. This catalogue only holds what an agent adds: authority, action, objective, memory, peers, radius and the record.

Which regime makes any of these controls mandatory?

This paper does not say, on purpose. Agentic Systems Under Regulation is the sibling that works out which duties reach an agent and by when, given that no agentic category exists in law. One control here does carry a date: disclosure and doubt, where Article 50 applies from August 2026.

Who should be reading this, and with what open?

Platform and application security engineers bounding an agent that already exists. It is a lookup document, so open it at a family rather than reading it through, and read it with the orchestrator and the tool gateway in front of you. Each entry names the component that enforces the control.

Get the whitepaper

Have it emailed to you.

Drop your details and we'll send Agentic Systems: An Engineering Reference straight to your inbox - no spam, unsubscribe anytime.

Download whitepaper
Next step

Your riskiest live agent, held against twenty-two controls in two weeks.

Two weeks on one agent already touching production systems: a per-agent principal, an argument-level allowlist on its riskiest tool, a kill switch drilled against the clock, and an architecture note on blast radius. Yours to keep. SECTION 7 - FAQ - 5 to 8 questions

Book an agent control review