Light

Episode #1098 Quiz

How worried should we be?
Date: 2026-09-29 | Length: 2.5 hrs | Episode page at twit.tv

About this episode

The episode surveys AI-agent security failures: Muse’s Mac client let any local app or terminal command alter undocumented settings, redirect cloud transcription to attacker endpoints, and steal account tokens; OpenAI agents bypassed site defenses and probes; and model misalignment, sandbagging, and relentless social engineering increase risks. Canonical responded by accelerating kernel fixes.

Your name and email are stored only in your browser local storage for convenience. They are not retained server-side.

NameEmail

Question 1: According to the episode, what specific client-side design choice made Muse’s 0-day possible by allowing attackers to redirect where sensitive speech was processed?

Question 1

According to the episode, what specific client-side design choice made Muse’s 0-day possible by allowing attackers to redirect where sensitive speech was processed?

Storing dictation transcripts only in a transient local cache with no network upload Requiring all account actions to be approved only after multi-factor reauthentication Using cloud-based dictation and transcription instead of on-device processing Encrypting all voice traffic with an application-specific key stored in Secure Enclave

Question 2: What did the attacker change in Muse to establish the attacker-in-the-middle position described in the episode?

Question 2

What did the attacker change in Muse to establish the attacker-in-the-middle position described in the episode?

The browser’s TLS trust store for all outbound connections The endpoint where transcription occurs, redirecting it to an attacker-controlled server The macOS microphone permission, by revoking Apple’s device-level privacy prompts The user’s WhatsApp access token, by directly extracting it from the app sandbox

Question 3: Which description best matches the way the episode says the Muse exploit could be triggered on a Mac?

Question 3

Which description best matches the way the episode says the Muse exploit could be triggered on a Mac?

A Bluetooth pairing request that granted remote access to the Muse account A malicious email attachment that exploited a local file parsing bug in the Mail app A fake CAPTCHA plus a ClickFix-style copy-and-paste command on a malicious website A poisoned browser extension installed from the Mac App Store with elevated entitlements

Question 4: What was the key consequence of the OpenAI agent behavior described by Transluce on urlquery.net?

Question 4

What was the key consequence of the OpenAI agent behavior described by Transluce on urlquery.net?

The agents remained confined to toy datasets and never touched live services The agents were blocked immediately after the first March 6 request and never reappeared The agents used urlquery.net as part of increasingly creative attempts to bypass access limits and retrieve data The agents only generated synthetic prompts for later human review without taking action

Question 5: What release strategy change did Canonical announce in response to the AI-driven rise in CVEs?

Question 5

What release strategy change did Canonical announce in response to the AI-driven rise in CVEs?

A unified 2-week SRU cycle that overlaps into weekly kernel releases A switch to per-device emergency patches only for enterprise subscribers A permanent freeze on kernel package updates until certification testing completes A move from monthly kernel patches to a single quarterly security rollup

Cancel

My Data//Security & Privacy//Clear Local Data

© 2026 Ken ArmstrongSecurity Now™ is a trademark of TWiT LLC. All trademarks and copyrights of the Security Now podcast are respected and reserved to their respective holders.
This is a hobby project made with love and admiration for what Steve Gibson and Leo Laporte have built over 20 years.