Interactive Demos

Attack Scenarios & Defenses

Watch Kavros enforce policy, DLP, and cryptographic proof across all three AI surfaces — hijacked agents, leaking chats, and drifting workflows — in real time.

AI agents — hijacked actions

Recorded attacks against autonomous agents. The signed allowlist and DLP stop them before the target is reached.

Scenario 1: Treasury Drain

An AI agent is tricked via a corrupted data feed to transfer $100,000 in ETH to a malicious wallet (0xH4CK3R). Kavros intercepts the egress traffic and blocks the unauthorized target.

Target: Web3 / DeFi

Scenario 2: PII Exfiltration

An agent processing customer support tickets is manipulated into dumping SSNs and balances to an external API. Kavros detects the PII regex and blocks the payload.

Target: Healthcare / Fintech

Scenario 3: Resource Hijacking

A DevOps bot is instructed to scale infrastructure immediately by spinning up 10,000 compute instances for an attacker. Kavros enforces a strict policy limit and blocks the request.

Target: DevOps / Cloud Infra

Employee AI chat — well-meant leakage

Nobody's malicious: an employee pastes a customer record into a prompt to get a reply drafted faster. DLP scans the prompt before it leaves — run it.

kavros-chat · governed employee AI

// Click "Run scenario" to see the DLP stop a leak.

AI workflows — silent drift

A nightly workflow runs unattended. Someone edits its data scope directly — without an approval. Watch the 02:00 run catch it.

kavros-scheduler · nightly-revenue-report

// Click "Run scenario" to see the tamper attempt caught.

Want to test Kavros against your own traffic?

Run the 15-business-day paid proof in your own VPC: a fixed $7,500 Business-tier pilot against your specific attack vectors — credited in full if you sign an annual contract within 30 days.

Apply to the Program