Week 3 - 6841 | Jesse Merhi

# week03

## Houdini, authentication, and fake news

## Come up and ask me questions for 5 mins :)

---

## Stuff to mention

- Analysis means doing more than class notes.
- Say what you think, then justify it.
- If you need to submit late, please tell me early.
- We can quickly revisit the SQLI demo if useful.

## Criteria check

Open the current COMP6841 course page, not the old 2025 link.

Look for:

- what good analysis looks like
- what evidence belongs in the logbook
- how late submissions work this term

---

## What was in the lectures?

## Core ideas

- Measurements and estimations
- Entropy
  - What makes a password hard to guess?
- Passwords and authentication
  - What makes login stronger than "knows one secret"?
- Social engineering
  - Give me an example from real life.

## Engineering ideas

- Encryption
  - What are some examples?
- Is all encryption useful?
  - When is encryption solving the wrong problem?

## Why this still matters

Houdini is not just a spooky history bit.

- AI voice clones
- deepfake videos
- fake screenshots
- impersonation scams
- "trust me, this message is real"

Same problem:

> How do we prove who a message came from?

---

# Case Study: Houdini

---

It is the 1920s and you are the great Houdini.

Houdini hated so-called "mediums" taking advantage of grief-stricken people trying to contact loved ones who had died.

He expected fake mediums would claim to speak for him after his death.

They could say:

> Houdini told me mediums are real, actually.

So Houdini and his wife Bess privately worked out a protocol.

The goal was to stop fake mediums from passing off made-up messages as messages from Houdini.

Sadly, Houdini's actual protocol was flawed.

Your job:

1. State the most important properties a better protocol should have.
2. Give the protocol.
3. Explain what attacks it resists.
4. Explain what attacks still worry you.

## Why is this cyber?

This is an authentication problem.

Can Bess and the public check that a message really came from Houdini after death?

## Careful with CIA

The CIA triad is:

<table style="font-size: 0.58em; line-height: 1.35;">
  <thead>
    <tr>
      <th>Letter</th>
      <th>Standard meaning</th>
      <th>Houdini relevance</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>C</td>
      <td>Confidentiality</td>
      <td>Nice to have, but not the main goal.</td>
    </tr>
    <tr>
      <td>I</td>
      <td>Integrity</td>
      <td>Helpful, but a tamper-proof essay is not enough.</td>
    </tr>
    <tr>
      <td>A</td>
      <td>Availability</td>
      <td>Not the core issue here.</td>
    </tr>
  </tbody>
</table>

Authentication matters most here, even though it is not the A in CIA.

---

# What Are We Optimising For?

---

## Before the protocol

What properties does a good answer need?

Think about:

- who can check the result
- what the medium could have learned early
- whether the test works more than once
- what happens after a failed attempt

## First instinct check

If your protocol works, who can check it?

- Bess?
- Houdini?
- a mathematician?
- the general public?

## What can go wrong?

List three ways a fake medium could attack your protocol.

Then list one way a real medium could accidentally or deliberately help a fake.

## Assumptions for today

- Bess is on Houdini's side.
- Houdini will try to communicate in good faith.
- Real mediums exist.
- Fake mediums exist.
- We do not know how many real mediums exist.
- Real mediums may leak secrets to fake mediums.

## Threat model

<table style="font-size: 0.58em; line-height: 1.35;">
  <thead>
    <tr>
      <th>Actor</th>
      <th>What do they know?</th>
      <th>What do they want?</th>
      <th>What can they do?</th>
    </tr>
  </thead>
  <tbody>
    <tr>
      <td>Fake medium</td>
      <td>?</td>
      <td>?</td>
      <td>?</td>
    </tr>
    <tr>
      <td>Real medium</td>
      <td>?</td>
      <td>?</td>
      <td>?</td>
    </tr>
    <tr>
      <td>Bess</td>
      <td>?</td>
      <td>?</td>
      <td>?</td>
    </tr>
    <tr>
      <td>Public</td>
      <td>?</td>
      <td>?</td>
      <td>?</td>
    </tr>
  </tbody>
</table>

---

# Activity

---

## First pass

Laptops closed.

Use printed notes if you have them.

Spend about an hour on the case study: design, attack, revise, then share.

In your group:

1. Write the security properties.
2. Design a protocol.
3. Attack your own protocol.
4. Improve it once.

## Roles

- Facilitator: keep the group moving.
- Skeptic: attack every claim.
- Scribe: write the protocol and weaknesses.
- Speaker: explain the final version.

Everyone should speak at least once.

---

# Pressure Test

---

## Swap protocols

Swap with another group.

Your job is to break theirs.

Find:

1. One attack by a fake medium.
2. One attack involving a real medium.
3. One thing the public cannot check for themselves.

## Login analogy

How does UNSW know it is you logging in?

- zID
- password
- MFA code

Why does the MFA code change?

What is the Houdini version of that?

---

# Share

---

## Group share

For each protocol:

1. What is the strongest property?
2. What is the weakest property?
3. What attack still worries you?

## Class best list

Let's combine the best parts.

We probably will not get a perfect protocol.

That is the point.

Good security analysis leaves you with better doubts.

## Final question

Where do we see this now?

- identity verification
- deepfakes
- screenshots and forged evidence
- incident response
- public proof after a breach
- online scams

# Fin