Hack, Pivot, Defend: A Live Account Takeover Face‑Off, With AI on Both Sides

On‑Demand Webinar

AI vs. AI, for real. In this recorded live session, an adversary uses agentic automation to probe your most sensitive journeys, while a defender responds in real time with edge telemetry, behavioral analytics, and intent‑based decisions. See the attacks, the signals they leak, and the precise counter‑moves that shut them down.

Runtime: ~45–60 minutes
Format: Demo + deep dive + audience Q&A

The Face‑Off

(recorded live)

An AI‑enabled adversary believes your defenses are predictable, bot controls, device fingerprints, digital identity checks. When they aren’t, he learns them and pivots. In this face‑off, Apollo.io’s Head of Fraud Prevention, Jon Ferrari, goes toe‑to‑toe with Darwinium’sBen Davi to expose how agentic automation actually works, and what still catches it today.

Across a series of increasingly complex ATO attempts, from synthetic sign‑ups and marketplace listing fraud to a phishing→session‑cookie hijack - you’ll see:

• Off‑the‑shelf browser agents (CDP/Playwright/Selenium wrappers) + LLM prompts driving reconnaissance and execution
• Real attack economics: why “week‑long” campaigns now execute in minutes at scale
• The defender’s playbook: edge instrumentation, behavioral/intent signals, and policy moves that block bad automation while letting good agents transact

You’ll leave with a practical checklist, model features you can ship this week (propagation delays, deterministic fill/typing cadence, shortcut navigation patterns), and a red‑team template to continually pressure‑test your flows.

What you’ll learn

• Real‑world tactics, MOs, and evolving AI‑driven attack patterns seen against Apollo.io and other merchants.
• What actually works (and what doesn’t) in AI‑native mitigation, including detection signals, edge instrumentation, and model‑based policies.
• Best‑practice approaches to anticipate the next wave — how to red‑team with agents and tighten your detection→mitigation feedback loop.

Plus, you’ll also see:

• Agentic AI in the wild: off‑the‑shelf browser agents + LLM prompts powering reconnaissance and execution.
• Phishing → session hijack via a quick reverse proxy, then operating as the victim.
• Detection signals: propagation delays, deterministic fill/typing cadence, shortcut navigation, and password/linkage quirks.
• How to allow good agents while blocking bad automation, without hurting UX.

Agenda

1. Welcome & context: Why agentic AI changes fraud economics.
2. Demo #1: Synthetic Account Factory: Agents scrape, learn local pricing, and post convincing listings with unique accounts/fingerprints.
3. Inside the agent: How LLM↔agent↔browser automation protocols work; where the telltale signals leak.
4. Demo #2: Phishing→Session Hijack: Build a minimal reverse proxy, capture cookies, and automate actions as the victim.
5. Detection & defense: What still catches agents today; what will catch them next quarter.
6. Fight fire with fire: Automating red‑team probes and accelerating the policy/ML feedback loop.
7. Q&A: Audience questions, rapid‑fire takeaways.

Key takeaways

• Agents ≠ bots-as-usual: Expect human‑like flows with rotating residential IPs, solver integrations, and rapidly improving mouse/typing emulation.
• Rules and CAPTCHA are not enough: Shift to signal‑dense telemetry and model‑based anomaly detection.
• Speed matters: Adversaries adapt in days; your feedback loop must be faster.
• Intent is king: When identity and device are ambiguous, journey‑level intent signals decide allow vs. challenge vs. deny.