muzzdang

Second Brain notes from YouTube & builds

Second Brain notes

AI Tools: Wtf is happening to the internet?

Watch on YouTube: https://www.youtube.com/watch?v=LD4a9IpoLMQ

Channel: Mo Bitar

What a Week of Cascading Outages Reveals About the State of Software Security

In a single week in late 2024, five critical U.S. systems — Microsoft Outlook, the FAA’s air-traffic network, Chase Bank, the NSA, and Apple — went dark. The cause wasn’t a coordinated cyberattack. It was routine patching. As Mo Bitar explains, organizations are being forced to deploy fixes at unprecedented speed because AI tools like Anthropic’s Mythos are uncovering vulnerabilities at a scale human teams can’t absorb. The result is a self-inflicted instability loop the creator calls “patchageddon.”

What the Video Covers

The video traces a six-day window where major infrastructure failed due to software updates, not intrusions. It connects three converging trends: AI-driven vulnerability discovery that compresses disclosure timelines, a massive increase in patch volume, and a deliberate reduction in pre-deployment testing. The argument is that we’re in a transition period — roughly 18 months — until AI can write secure code from the first commit (what the video calls “v1.0”). Until then, the same companies finding the holes are also selling the tools to fix them, creating a trillion-dollar remediation market.

Step-by-Step: How the Cycle Works

  1. AI finds flaws at scale. Anthropic’s Mythos reportedly identified holes in classified systems in hours. The tool was shared with a small group of large corporations for early remediation.
  2. Disclosure windows shrink. Google cut its vendor fix window from 90 to 60 days; Anthropic set a 30-day deadline for critical vulnerabilities.
  3. Patch volume explodes. Microsoft’s November 2024 Patch Tuesday shipped 1,094 fixes — up from 63 a year earlier. Cisco ThousandEyes data shows U.S. network outages rising from 65 per week to 243 per week.
  4. Testing is skipped. Both Anthropic and JPMorgan urged shorter patch testing and deployment timelines. Trail of Bits found that roughly 1 in 6 security patches (21 of 130) broke something else.
  5. Outages cascade. Chase’s outage was traced to a Monday software release on-premise. The FAA saw a single new router in El Dorado, AR generating 1,000 hits every 4–5 seconds. Microsoft Outlook’s outage affected 74 countries with an estimated $7B productivity loss.
  6. Organizations throw people at the problem. Bank of America hired 1,000 contractors for patching work. The NSA took down its own network while patching.

Key Numbers and Tools

  • Microsoft Patch Tuesday: 63 fixes (Nov 2023) → 1,094 fixes (Nov 2024)
  • U.S. network outages: 65/week (2023) → 243/week (2024) per Cisco ThousandEyes
  • Microsoft Outlook outage: 74 countries, ~$7B productivity loss estimate
  • FAA router anomaly: 1,000 hits per 4–5 seconds from one new router
  • Google disclosure window: 90 days → 60 days
  • Anthropic critical-hole deadline: 90 days → 30 days
  • Trail of Bits: 21 of 130 security patches (~16%) broke something else
  • Bank of America: 1,000 contractors hired for patching
  • Enjoy.dev: free trial at enjoy.dev; runs on the user’s existing Claude or ChatGPT plan

We basically shot ourselves cleaning our own gun five times in 6 days

Patchageddon: there is no last patch. It’s patches all the way down

Test less, ship faster. That was their plan. And this is a bank we’re talking about

Those are Russian roulette odds, man

You have an AI that’s able to read like every line of code ever written. It reads it once and says, ‘Hey, this all needs to change.’

Caveats and Hype Check

The creator plugs his own product, Enjoy.dev, heavily in the final minutes — an affiliate/self-promotion signal. The “trillion-dollar business model” claim is a speculative revenue projection, not a reported figure. The 18-month timeline to “v1.0” secure code comes from a single Anthropic employee (Jason Clinton, deputy head of security) and is an unproven prediction. The $7B Outlook productivity loss assumes all Outlook time is productive, which inflates the estimate. Treat these as directional indicators, not verified forecasts.

Concrete Next Actions

  • Monitor your organization’s patch cadence: expect more frequent, larger patch batches with shorter test windows.
  • Budget for rollback capacity: Trail of Bits data shows 1 in 6 security patches breaks something else.
  • Evaluate AI-assisted code review tools (e.g., Claude, GitHub Copilot) to catch vulnerabilities pre-deployment.
  • Track vendor disclosure deadlines: Google 60 days, Anthropic 30 days for critical holes.
  • Try Enjoy.dev free if you create video or software content — it automates editing, asset generation, thumbnails, and short-form cuts using your existing AI plan.

The Bottom Line

The internet’s current instability isn’t a bug — it’s a structural consequence of AI accelerating vulnerability discovery faster than human processes can safely remediate. The next 18 months will likely bring more outages, not fewer, as organizations compress testing to meet shrinking disclosure windows. The strategic move isn’t to patch faster; it’s to invest in tooling that shifts security left so the patches become unnecessary. Until then, expect smoke signals.