Making AI agents secure and insurable

Your guardrails don’t know what your agent is supposed to do.

JANUS recovers the SPEC your agent should follow, tests it, and enforces it at runtime.

Book a demo See the platform
CRPO / SMU Spin-offSingapore · SaaS or on-premises

01 The risk

Three risks behind AI agent incidents. JANUS covers all three.

80%

Supply-chain risk

Vulnerable or malicious models, packages, MCP servers and skills.

up 4× since 2023

→ JANUS Audit · AI-BOM

74%

Traditional security risk

Flaws in agents and platforms: injection, auth, sandbox escape.

up 5.6× since 2023

→ JANUS Pentest
→ JANUS Runtime Monitor

20%

Business-rule risk

Rules set in prompts aren’t enforced like code.

up 11× since 2023

→ JANUS Audit · SPEC Recovery
→ JANUS Runtime Monitor

Share of AI agent incidents, 2023–2025. Incidents can overlap. Source: SMU Lab Research.

02 The platform

JANUS

Recover.
Attack.
Enforce.

One engine, three products. Each one starts from the same question: what is this agent supposed to do?

Explore the platform
Recover01

JANUS Audit

Recovers the business rules your agent should follow, from its code and prompts, and tests every one.

SPEC RecoveryAI-BOM
Attack02

JANUS Pentest

Simulated attacks on agents and their AI components. Evidence, not a score.

Prompt injectionTool hijackingMCP / A2A abuse
Enforce03

JANUS Runtime Monitor Early Access

Enforces the SPEC and guard rules on every live action.

AllowBlockEscalateLog

03 How it works

Before go-live, JANUS writes down what your agent should do. In production, it holds the agent to it.

RECOVERJANUS AuditSPEC Recovery · AI-BOMreadsYOUR AGENTcode · prompts · tools · models · MCPattacksATTACKJANUS Pentestsimulated attacksbusiness rules · AI-BOMguard rulesSPEC & POLICYbusiness rules · AI-BOM · guard rulesBEFORE GO-LIVEIN PRODUCTIONenforced at runtimeyour agent (live)LLM & tool callsENFORCEJANUS Runtime Monitorallow · block · escalate · logEARLY ACCESS
Recover

JANUS Audit

Reads your agent — code, prompts, tools, models, MCP — and recovers business rules and an AI-BOM.

Attack

JANUS Pentest

Simulated attacks on the agent produce guard rules.

SPEC & POLICYbusiness rules · AI-BOM · guard rules
Enforce

JANUS Runtime Monitor Early Access

Sits between your live agent and its LLM & tool calls: allow · block · escalate · log.

04 For insurers

The same JANUS engine is built to support insurers in underwriting AI agent risk.

A recovered SPEC, an AI-BOM and attack evidence describe an agent the way an underwriter needs to see it: what it should do, what it is made of, and how it fails.

JANUS for insurers

05 Provenance

CRPO / SMU
Spin-off

CRPO
Research centre under CSA.
SMU
Singapore Management University.
  1. Government01

    Deployed at a national government agency

  2. Insurance02

    Channel partnership with a leading global insurance firm

  3. Academia03

    Being adopted by a leading university

  4. Deployment04

    Available as SaaS or on‑premises

Underpinned by peer-reviewed research at ICSE, IEEE S&P, USENIX Security, NDSS, FSE

Let’s look at your agent.

Book a demo

or write to support@agentsure.tech