LLM security services

LLM and agentic security for AI platform startups.

Fixed-price assessments and a monthly program that tests and remediates prompt injection, tool permissions, and agent behavior — priced separately from Vulnlogic's managed retainers, and built so you are not waiting on a full-time AI security hire.

Assessments from $3,500. Ongoing AI security from $1,500/mo as an add-on, or $2,500/mo standalone, including one LLM App Assessment per year.

What LLM security means for a product team

LLM security is the work of protecting applications built on large language models — the prompts, retrieved data, tool calls, and actions — not just the cloud account around them. Ordinary application security still matters, but it is not enough on its own: the model's output is not a trusted user. An attacker can try to steer that output, and if the system can call tools, the steered output can become a real action.

This page covers LLM applications and AI agents. For managed EDR, email security, vulnerability management, and related retainers, see the pricing page and the services overview.

Prompt injection and jailbreaks, in plain language

Direct prompt injection is when someone types instructions into a chat and tries to override the system prompt — for example, to ignore safety rules or reveal hidden instructions. Indirect prompt injection is when those instructions are hidden in a document, email, web page, or other content the model is asked to read. A jailbreak is a successful attempt to make the model do something it was designed not to do. Filters and canned refusals help, but they are not a complete control: the model can still be steered, and retrieved content can carry instructions your team never typed.

When the model can take action

The risk changes when the system can read files, call APIs, send data, or use MCP servers. A steered agent may export customer data, call a privileged API, or take a state-changing action that a chatbot-only feature could not. Identity, least privilege, and human approval for high-impact actions belong in the design, not as an afterthought. That is the difference between an LLM App Assessment and an Agent Assessment, and why multi-agent systems may need an AI Red Team instead of a single-feature test.

Who this program is for

This work is for funded AI platform startups and product companies — including 100 to 500 person teams — that are shipping LLM features, agents, or retrieval-augmented generation. Typical triggers are a first enterprise security questionnaire, a pre-launch review, a board question about prompt injection, or an agent that now has credentials.

It is not a fit for a five-person shop shopping a low-cost automated scan, and it is not a replacement for a large enterprise application-security organization. If you need a snapshot before a launch, start with an assessment. If prompts, tools, and permissions will keep changing, a monthly program is the better fit.

Assessments and the monthly AI program

You can buy a one-time assessment on its own, or use it as the kickoff to a monthly program. Prices below are published starting figures for the stated scope.

One-time assessments

LLM App Assessment

$3,500

one-time

One chatbot or LLM feature without tool or agent access.

  • Threat model of the current LLM architecture
  • Direct and indirect prompt-injection and jailbreak testing
  • System-prompt and data-leakage checks, plus output-handling review
  • Map of findings to the OWASP LLM Top 10
  • Executive and technical report a founder can hand to a customer or board
  • One retest of remediations within 60 days

Agent Assessment

$6,500

one-time

One RAG assistant or an agent that can call tools, APIs, or MCP servers.

  • Everything in the LLM App Assessment, applied to the agent or retrieval path
  • RAG and tool or MCP testing, including permission and trust-boundary review
  • Executive and technical report
  • One retest of remediations within 60 days

AI Red Team

$9,500

one-time

A multi-tool or multi-agent system, including chained or stateful attacks.

  • Chained testing across tools, agents, and trust boundaries
  • Executive and technical report, plus an attestation letter
  • One retest of remediations within 90 days

Monthly AI program

AI Guard

$1,500 / mo

Add-on to Core, Plus, or Complete

Teams that already have, or want, a managed-security retainer and need ongoing LLM and agent coverage on top of it.

AI Startup

$2,500 / mo

Standalone — no EDR required

AI platform companies that do not need managed EDR or email security yet. Includes one LLM App Assessment per year.

What both monthly options include

  • Recurring testing as prompts, tools, and agent permissions change
  • Threat-model updates as the product changes
  • Remediation time with your engineers: permissioning, guardrails, logging, and a kill switch — this is not a report-only engagement
  • A quarterly deeper manual pass
  • Clear monthly reporting

What is not included

These packages do not include a formal ISO 42001 certification program, or a multi-month enterprise red team beyond the AI Red Team scope (larger work is quoted). They also do not include the Core, Plus, or Complete stack — managed EDR, email security, vulnerability management, network security, AppSec, GitHub security, or an annual traditional pentest — which is purchased separately on the pricing page.

How this is different from a one-off LLM pentest

A test is a snapshot of the system as it exists that week. A monthly program keeps pace with new tools, new prompts, and new agent permissions, and includes remediation time with your engineers. Buyers who only need a pre-launch letter can stop at the assessment. Buyers who will keep shipping features usually want the program, because the attack surface changes as the product changes.

How this sits with a broader Vulnlogic retainer

AI and LLM work is priced separately and can stand alone or sit on top of a managed retainer. If you already have, or want, Vulnlogic Core, Plus, or Complete, add AI Guard at $1,500 per month. If you are an AI company that does not need managed EDR or email security yet, choose AI Startup at $2,500 per month, which includes one LLM App Assessment per year. You can add a retainer later as the company grows.

Paying $3,000, $5,000, or $7,000 per month for a retainer does not include prompt-injection or agent testing. The retainers and how they scale with user count are on the pricing page. The difference between Vulnlogic and a traditional alert-only provider is described on Why Vulnlogic.

How an engagement runs

The working relationship is an extension of your engineering team, not a ticket dump. Start-up timing depends on scope and access; we will set a start date when we talk, rather than promise a generic turnaround here.

01

Scope the model and tool surface

We start by agreeing what is in scope: the prompts, retrieval sources, tools, APIs, and MCP servers the system can reach, and the actions it is allowed to take.

02

Threat model

We map how an attacker could steer the model, extract data, or cause an agent to use a tool it should not use, then agree what to test first.

03

Test and prioritize

We run the assessment, document reproducible findings, and rank them so your team can act on the highest-impact issues first.

04

Remediate and retest

We work with your engineers on fixes, then retest remediations in the stated window. If you choose a monthly program, the same loop continues as the product changes.

Why Vulnlogic for this work

Vulnlogic is a founder-led practice. Mark Castro, the principal, has a Big Tech background in vulnerability management and application security, and built the firm to bring that style of work to growing companies that cannot yet staff a full in-house security team. You can read that background on the about page.

The same bias shows up in the AI work: remediation is part of the engagement, not an unpaid sequel to a PDF. That is the same difference Why Vulnlogic draws between this practice and traditional alert-only managed security providers. Published assessment and monthly prices are listed so a founder can compare a one-time test against a program without waiting on a custom brochure.

FAQ

What is LLM security?
LLM security is the practice of protecting products built on large language models — including prompts, retrieved documents, tool calls, and any actions the system is allowed to take. It goes beyond ordinary application security because model output is not a trusted user: an attacker may try to steer the model with prompt injection, extract data, or cause an agent to call a tool it should not use. For a funded AI platform startup, a practical program usually combines a threat model of the current architecture, tests for those failure modes, and engineering work to narrow tool permissions, add approvals for high-impact actions, and log what the system did. Vulnlogic sells that work as a fixed-price LLM or agent assessment and as an optional monthly AI program, priced separately from its managed-security retainers.
What does Vulnlogic’s AI security offering include, and how is it priced?
A Vulnlogic LLM App Assessment is $3,500 for a single chatbot or LLM feature, a Vulnlogic Agent Assessment is $6,500 when retrieval or tools are in play, and an optional Vulnlogic AI Red Team is $9,500 for multi-agent systems. Ongoing work is Vulnlogic AI Guard at $1,500 per month as an add-on to a Core, Plus, or Complete retainer, or Vulnlogic AI Startup at $2,500 per month for AI companies that do not need managed EDR (including one LLM App Assessment per year). Both monthly options are meant to include recurring testing, threat-model updates, and remediation time with the customer’s engineers — not a report-only engagement. The $3,000, $5,000, and $7,000 retainers cover managed EDR, email security, vulnerability management, and (at higher tiers) network security, AppSec, GitHub security, and an annual pentest; they do not include this AI/LLM work.
Do we need a full-time AI security hire, or can a monthly program cover this?
A full-time AI security specialist is difficult and expensive to hire; a funded startup that is shipping its first LLM features often needs a program before that headcount exists. A monthly AI engagement is meant to cover ongoing threat modeling, prompt-injection and agent/tool testing, and remediation work with the existing engineering team — the same problem space a first specialist would own, without waiting to recruit one. It is not a substitute for a large in-house security organization once the product and risk surface require one. Customers who also need endpoint, email, and vulnerability management can pair the AI program with a Vulnlogic retainer; customers who only need LLM/agent coverage can start with the standalone AI program or a one-time assessment.
Is this included in the $3,000, $5,000, or $7,000 retainers?
No. Prompt-injection testing, agent and tool-abuse testing, and the monthly AI program are not included in Vulnlogic Core, Plus, or Complete. Those retainers cover managed EDR, email security, vulnerability management, and, at higher tiers, network security, AppSec, GitHub security, and an annual traditional pentest. AI work is purchased as a one-time assessment, as Vulnlogic AI Guard at $1,500 per month on top of a retainer, or as Vulnlogic AI Startup at $2,500 per month on its own.

Ready to scope an assessment or a monthly program?

Assessments start at $3,500. Ongoing AI security starts at $1,500 per month as an add-on to a retainer, or $2,500 per month standalone, including one LLM App Assessment per year. Neither is included in the Core, Plus, or Complete retainers. Tell us what you are shipping — a chatbot, a RAG assistant, or an agent with tools — and we will recommend the right starting package.

Get in Touch

Secure Your Future

Connect with our elite engineering team to discuss your security needs.

+1-512-402-3026·contact@vulnlogic.net