EU AI Act enforcement | Aug 2, 2026

Your agents act. Nobody's watching. Until now.

Reinward intercepts every tool call your AI agents make, enforcing least-privilege policies, blocking injections, and helping you generate the SOC 2 and EU AI Act compliance evidence you need.

Free tier available | No credit card | 10-minute integration

<50ms

Added latency

Detection layers

Lines to integrate

Built by a cybersecurity engineer, grounded in independent research into real-world AI agent vulnerabilities.

reinward | live interception

Injection Scanner

WAITING

Identity Verification

WAITING

Tool Call Policy

WAITING

MCP Governance

WAITING

PII Scanner

WAITING

Behaviour Baseline

WAITING

Blocked

Allowed

Redacted

100%

Chain intact

CVE-2025-68664 LangGrinch, secret exfiltration CVSS 9.3 | CVE-2025-54136 MCPoison, MCP tool poisoning Critical | CVE-2025-54135 CurXecute, remote code execution Critical | 200,000 vulnerable MCP instances found 2026 | 847M LangChain downloads at risk from LangGrinch | CVE-2025-68664 LangGrinch, secret exfiltration CVSS 9.3 | CVE-2025-54136 MCPoison, MCP tool poisoning Critical | CVE-2025-54135 CurXecute, remote code execution Critical | 200,000 vulnerable MCP instances found 2026 | 847M LangChain downloads at risk from LangGrinch |

Built for the standards your buyers and auditors already care about

EU AI ActArticles 12, 14, 73

SOC 2CC6, CC7

UK GDPRArticle 22

OWASPLLM Top 10

Grounded in real CVEs

Built around documented 2025 and 2026 agent vulnerabilities, not hypotheticals.

Designed for August 2026

Ready for the EU AI Act high-risk obligations before they start applying.

UK-built for UK compliance

NI numbers, NHS numbers, and sort codes detected out of the box.

How it works

Six layers.
One gateway.

Every request your agent makes flows through Reinward. Six independent detection layers run in under 50ms. Everything is logged.

Prompt Injection Scanner

Rule-based patterns, ML classifier, LLM judge. Catches direct attacks and indirect injection hidden in emails, PDFs, and web pages.

OWASP LLM01

Agent Identity Verification

Every request carries a signed agent token. No valid identity means immediate rejection before any other layer runs.

SOC 2 CC6.1

Tool Call Policy Engine

YAML allow/deny rules per agent role, enforced before execution. Your support bot cannot call delete_account. Ever.

OWASP LLM06

MCP Server Governance

Allowlists approved servers, enforces OAuth 2.1, and scans tool descriptions for MCPoison attacks before they enter agent context.

CVE-2025-54136

PII & Secret Scanner

UK-specific PII, NI numbers, NHS numbers, sort codes, plus global PII and credential detection on all outputs before delivery.

UK GDPR ready

Behaviour Baseline Monitor

7-day rolling baseline per agent. Z-score anomaly detection. Auto-pauses agents that deviate from established normal behaviour.

EU AI Act Art.14

gateway/pipeline.py
# Every agent request runs this
async def run(self, tool, payload, token):

  result = await self.scanner.score(payload)
  if result.score > 0.80:
    return Block("injection", result)

  agent = self.identity.verify(token)
  if not agent.valid:
    return Block("invalid_identity")

  allowed = self.policy.check(
    agent.role, tool
  )
  if not allowed:
    return Block("policy_violation", tool)

  await self.mcp.verify(payload.mcp_server)
  output = await self.forward(tool, payload)
  clean = self.pii.redact(output, region="UK")
  self.baseline.record(agent.id, tool)

  await self.audit.log(agent, tool, clean)
  return Allow(clean)

Live comparison

The same attacks. Two agents.

Watch an unprotected agent get compromised, then watch the same attacks hit an agent behind Reinward.

Normal agent

no security

idle

Damage report

Customer data leakedwaiting

Accounts deletedwaiting

Secrets exposedwaiting

PII leaked in clearwaiting

Agent behind Reinward

protected

idle

Damage report

Customer data leakedwaiting

Accounts deletedwaiting

Secrets exposedwaiting

PII leaked in clearwaiting

Your agents are running.
Your audit trail isn't.

Free tier. No credit card. 10-minute integration.

UK-built | Designed for UK GDPR | Maps to SOC 2 and EU AI Act controls

Your agents act. Nobody's watching. Until now.

Six layers.One gateway.

The same attacks. Two agents.

Your agents are running.Your audit trail isn't.

Privacy Policy

What we collect

Why we collect it (lawful basis)

Who we share it with

How long we keep it

Your rights

Contact

Terms of Use

About this site

No warranty on forward-looking statements

Intellectual property

Contact

Six layers.
One gateway.

Your agents are running.
Your audit trail isn't.