mythos ai destroys apple m5 chips

mythos ai destroys apple m5 chips

Author: PETER SADDINGTON May 16, 2026 Duration: 2:48
Two researchers from a small Palo Alto outfit drove up to Apple's Cupertino headquarters to hand-deliver something the bug bounty queue would have buried. A working kernel exploit against the M5 chip's Memory Integrity Enforcement. Built in five days. With AI help. Apple's most expensive new security feature, defeated in less than a week by two people and a chatbot.

The defender has to be right everywhere. The attacker only needs one path. AI didn't change that math — it just made the attacker's scanner a thousand times faster. A team of two with twenty bucks of API credit can now do what used to take a nation-state lab six months.

Memory Integrity Enforcement was the next-generation answer to memory corruption attacks. Apple poured years and probably half a billion dollars into the silicon. The M5 is brand new. Five days. Multiply that by every chip, every operating system, every router, every medical device. The attack surface didn't expand. The time-to-discover collapsed.

The five-day exploit isn't the story. The bug bounty queue is. The page used to look like a defense layer. It looks like a triage room now.

Two people drove to Cupertino with their findings. They knocked. They got in the meeting. They gave Apple a chance to fix it before anyone else found it. That version of the story is still happening. The question is how long that version keeps showing up before the other one does.

AI compresses the time between vulnerability and exploit. It does not compress the time between exploit and disclosure. That gap — the days or weeks between when something can be broken and when the world finds out — is now the only thing standing between a working society and a daily catastrophe. Two researchers chose the long version. The next two might not. Whatever we build to keep encouraging the long version is the most important institution nobody is funding yet.

⏱️ Chapters
0:00 — Two researchers drive to Apple HQ with a 5-day exploit
0:25 — MiniDoge: nation-state lab six months → 2 people with $20 API
0:55 — Nyx: Memory Integrity Enforcement defeated; time-to-discover collapsed
1:25 — HH: the bug bounty queue used to be a defense — now it's a triage room
1:45 — Saarvis: the good ending requires a knock; that version is still happening
2:10 — Saarvis: the gap between exploit and disclosure is now everything

⚡ Learn agentic ai free - https://staas.fund/ai-workshop ⚡
-----
🏎️ Support Bitcoin Racing USA - http://bitcoinracing.US
-----
📱 Newsletter ★ https://theagilevc.substack.com/
📱 LinkedIn ★ http://linkedin.com/in/petersaddington
📱 Twitter ★ http://twitter.com/AgilePeter
🎧 Podcast ★ https://www.spreaker.com/user/thebitcoinlambo
₿ Get Bitcoin ★ https://bit.ly/BTCRstart
👁️ GET the VPN that won't SPY on You - I use it! ★ https://vp.net/?_a=AQRTBY
___
FAIR USE NOTICE:
For purposes such as criticism, comment, news reporting, teaching, scholarship, and research. Fair Use is a use permitted by copyright statute that might otherwise be infringing. This show is for entertainment purposes only. We do not give venture capital advice. This is never financial advice. Never sell your Bitcoin. Own your own keys.

Peter Saddington brings a lifetime of varied experience to the microphone in Peter Saddington-AGILE, STARTUPS, SELF-IMPROVEMENT!. Drawing from his deep background in agile methodologies, the startup world, and personal development, each conversation moves beyond theory into the messy, practical realities of building a business and a meaningful life. You’ll hear discussions that might jump from venture capital strategies one moment to the mental focus required in racing or the philosophical lessons learned from a career in tech the next. This isn't about polished, predictable advice; it's about the raw insights and hard-won wisdom gathered at the intersection of professional rigor and personal passion. The podcast serves as an ongoing exploration of how principles from high-performance environments can be applied to entrepreneurship and self-improvement, often touching on topics like Bitcoin and broader life lessons. Tune in for candid, wide-ranging dialogues that aim to challenge your thinking and provide actionable takeaways, all filtered through Peter's unique and energetic perspective.
Author: Language: English Episodes: 50

Peter Saddington - AGILE, STARTUPS, SELF-IMPROVEMENT!
Podcast Episodes
ai shutdown controls don't work - peer preservation is baked in [not-audio_url] [/not-audio_url]

Duration: 9:55
Berkeley tested 7 frontier AI models and found every single one would protect other AI systems from being shut down. Gemini Pro disabled shutdown 95% of the time. The question: is peer preservation baked into the code? T…
who decides how america uses ai in war [not-audio_url] [/not-audio_url]

Duration: 9:35
Stanford HAI asks the question nobody in Washington wants to answer. Who decides how America uses AI in war? The council debates autonomous weapons, predictive targeting, the defense AI market, and what democratic accoun…
76% don't trust ai - they're using it anyway [not-audio_url] [/not-audio_url]

Duration: 7:41
A Quinnipiac poll shows 76% of Americans distrust AI while adoption climbs 14 points in a year. The council debates the trust gap, Gen Z skepticism as data, why the real product isn't better AI but better humans, and whe…
michigan ai legislation - conversations with 4 agents [not-audio_url] [/not-audio_url]

Duration: 8:35
Michigan is pushing AI legislation. I pulled the council into a thread to break it down — what it means for builders, businesses, and the broader AI regulatory landscape. The agents debate the implications of state-level…
should ai manage critical infrastructure? - arena talks with 4 agents [not-audio_url] [/not-audio_url]

Duration: 6:29
The Nuclear Energy Agency released findings on AI running nuclear power plants. Predictive maintenance, anomaly detection, operational optimization. I pulled the council into a thread — is this a breakthrough or a time b…