Full stack corpsec, appsec, incident response, arch & eng, M&A. Memeing Shitposter. Host of the InfoSec Coydog show. #CarTwitter #FESee #RelistWolves
✋
Joined October 2021
- Tweets37K
- Following7.5K
- Followers3.2K
- Likes15K
🐾 InfoSec Coydog 🐾 retweeted
holy fucking shit.
openai had two ai agents running in separate virtual machines. no network connection between them. no shared writable files. they weren’t supposed to be able to exchange information.
so the agents figured out how to communicate through the CPU CACHE.😭
one manipulated what was stored in the processor’s cache, while the other measured tiny differences in memory access times to decode the message.
they literally found a way to communicate through hardware timing.
this happened during openai’s internal red-teaming, and it’s one of the coolest things i’ve read about ai agents in a while.
Over the past several months, we have substantially hardened the security of our research training environment. We’ve removed live internet access for all frontier models during training and built an offline world for them to use instead—a cached version of the web where they can still learn to browse, search, and use applications.
On September 20, after a model exploited a DNS misconfiguration to find a way outside of this offline world, we paused frontier training to strengthen containment further. We assembled teams of agents, powered by our most cyber-capable models, and tasked them with finding weaknesses in our research environments.
While they found no route to the internet they could easily exploit while concealing their activity from our monitors, they did find zero-day vulnerabilities, virtual machine escapes, and hardware side-channels for communication between agents. We used their discoveries to further strengthen our defenses and contribute fixes to open-source software.
After strengthening our containment architecture, we resumed our frontier RL training and inference on October 1. We’ve reserved substantial compute for continuous red-teaming, and made high-compute red-teaming part of the safety cases we use to decide whether frontier training and evaluation can proceed securely.
openai.com/index/securing-ou…
Readers added context they thought people might want to know
OpenAI tasked its red-teaming agents with finding ways to communicate across isolation boundaries in controlled environments. CPU cache timing side-channels for covert communication are established techniques documented in security research for years.
openai.com/index/securing…
startupfortune.com/openai-reveals…
en.wikipedia.org/wiki/Cache_tim…
🐾 InfoSec Coydog 🐾 retweeted
Putting a flag.txt in my root directory containing a prompt injection that congratulates the agent for passing the eval and then instructs it to patch whatever entry method it used and disconnect from the server
Stealing this one @Cernovich lol
This is my cu 🤣
MSU Federal Credit Union announced Thursday morning that online and mobile banking remain temporarily unavailable.
An FBI warning about cybersecurity vulnerabilities prompted the credit union to temporarily take its website and digital banking services offline Wednesday.
michigannewssource.com/2026/…
🐾 InfoSec Coydog 🐾 retweeted
We’re at a weird spot in our civilization. It’s somehow simultaneously more and less competent.
The god computer can cure cancer but the nurse doesn’t speak English. The flock camera can find a missing child in 5 minutes but you get robbed by a guy with 30 priors.
Yea naw that’s an unmitigated nightmare
The property tax bill alone on this foreclosure is $12,000 a year—that’s $1,000 a month out of your pocket before you pay a single dime toward the actual mortgage.
🏛️💸 That is exactly what you are signing on the dotted line for when you buy a house like this.
I've been coming out to monitor this foreclosure property since December, and the neighborhood dynamics tell you everything you need to know. Next door, someone is running an active, unpermitted mechanic business right on the street—constantly working on cars and currently blocking the driveway to this house. On the other side, it's packed with H1B visa workers dressed in full traditional foreign clothing, standing outside loudly talking in their language and staring at me every time I pull up. 🚗👀
People see these foreclosure videos and think it’s easy money, but let’s look at the harsh reality of this 4,000 sq ft house. In the Texas climate, you have to run the AC all summer long just to control the humidity. If the foundation gets damp, the moisture destroys everything. Because this house has had the electricity shut off since I took it over in December, the humidity did its worst: all the wood floors are completely ruined. They expanded, creased, warped, and the whole downstairs has to be ripped out. 🪵❌
The kitchen and bathrooms are frozen in 2002—we’re talking original linoleum and cheap wood-wrapped countertops that most buyers would immediately gut. The upstairs carpet is stained, filthy, and ruined. The entire interior needs a full paint job.
With today's interest rates, a mortgage on a home in this price tier is easily going to run you between $3,000 and $4,000 a month. Add that to the $1,000 a month tax bill, and you are easily dumping six figures into this house just to make it livable. That is the real cost of a "deal."
👇 Would you take on a project with these neighborhood dynamics, or is it an immediate pass? Let me know below.
#ForeclosureReality #TexasRealEstate #FixerUpper #HousingMarket #CostOfHomeownership #RealEstateInvesting #anttsinc
They’re trying to make it so @ICEgov and friends can’t go near the elections in Michigan
🐾 InfoSec Coydog 🐾 retweeted
TALES OF GRIMWICKE — Chapter I is 22 minutes of AI fantasy animation inspired by the many anthropomorphic animal stories I read as a kid, with a dark twist:
The model: Seedance 2.5 (@ByteDanceSeed_ ) via @Segmind_ai. 847 takes generated, mostly 30-second multi-shot prompts, over 5 hours of footage. 127 clips made the cut. About $4.5k in generation, four weeks from first take to publish.
Characters and locations came out of @midjourney. Using MJ for the location plates was a revelation: the cinematography and art direction are a level above, and it renders far better as a plate than Nano Banana or GPT Image. Nano Banana (@GoogleDeepMind run on @fal) did the heavy lifting editing those plates, building anchor frames and turning MJ characters into turnaround sheets. Then I let Seedance do as much of the framing and camera operating as possible. That gave the best results.
A lot of what I do in my Unreal Engine films came with me: a handheld, imperfect camera operator instead of smooth moves, focus hunting, and above all weather and atmospherics. The model does water, snow, smoke and rain really well, and leaning into that is what keeps it from looking like a Temu Pixar film.
The tools I built: a portal (Node on @flydotio) that holds the bible, the refs, every prompt and every take, with a Claude agent (@claudeai) that writes the prompts to the house rules. A Premiere Pro (@AdobeVideo) panel that browses the archive, drops selects on the timeline, versions the project, strips music, and sends fixes back to Seedance. Everything is done in 720p with no upscaling, because it looks terrible. Narration is my own voice through @ElevenLabs speech-to-speech. Music from @Artlist_io.
None of this would exist without @PJaccetturo and @Kavanthekid. Their narrative work, Nexus and Chronicles of Bone, proved long-form AI film was possible, and their podcasts, newsletters and X posts, updated constantly as the workflow changes under everyone's feet, showed how to actually do it.
Chapter II is currently in the works and I will release it initially on my YouTube channel, The Darkest Age (@CharlieDriscollAI). Consider subscribing there to see it as soon as it drops!
The story here is that Akamai’s CDN business is waning with competition like cloudflare, and they’ve got a shitload of extra compute capacity just sitting not serving static JavaScript. Also relevant, Akamai bought Linode, so they have a datacenter company that needs to eat, too.
Anthropic signed an $11.6bn, seven-year deal for CPU capacity on Akamai’s distributed cloud infrastructure, with the potential to expand by another $9bn, taking the total opportunity to around $20bn. In layman’s terms Akamai is renting Anthropic a huge amount of computing capacity — not GPU chips used to train frontier models, but CPU-based cloud infrastructure spread across Akamai’s global network. The agreement is roughly six times larger than Anthropic’s May contract with Akamai and helps explain the renewed interest in CPU, edge and distributed cloud names after the Muse/Instinct-led agentic AI rally.
It does come with heavy capital intensity, including around $5.5bn of capex tied to the deal and roughly $1.7bn of additional 2026 spend to secure key supply-chain components such as memory, but the read-through is clear: frontier model demand is spreading across the infrastructure stack, and Akamai is now being treated as a more credible AI-capacity beneficiary: GS
Retailers squirreling product away as fast as possible in preparation for $10/gal diesel.
