
Startup mechanic. Previously CEO of @Macrofab and Co-Founder of @Alertlogic.
Houston, TX
Joined February 2009
- Tweets36.8K
- Following1.4K
- Followers5.6K
- Likes5.3K
Misha G. retweeted
When even the guy who once worked at Enron says WTF...
Billions of dollars’ worth of contracts predicting oil would fall and the S&P 500 would rise were placed about 15 minutes before Trump announced a U-turn on Iran in a post on Truth Social. CFTC staff began to investigate-- then were told to lay off. bloomberg.com/news/features/…
Misha G. retweeted
We’re live in a bubble. Everyone we follow builds stuff, so we start believing everyone wants to. They don’t.
Most people dont give a crap about building software. In fact, most people hates computers.
They just want the dishwasher to work and the app to open. Nobody’s spending their weekend chatting with Claude on their phone. They’re living their lives, and they want tech to stop getting in the way.
Misha G. retweeted
Houston is the only major city in the US that has no zoning. That means we don't ration housing, making housing incredibly cheap. When you make housing cheap, every service related expense also becomes cheap. It's a natural inflation deterrent.
Misha G. retweeted
I'm a journalist. And I happen to be Jewish. On Yom Kippur I went to a synagogue service in Toronto and was met by protesters trying to taunt and intimidate me from entering the building. 1/5
Misha G. retweeted
Two guys replied to my post, both of them have started highly successful companies, and really don’t need to work for money.
Yet they’re out here learning and growing with these new tools. Pushing themselves to be better by doing the work.
HUGE Lesson here.
xitter.cf/rossiadam/status/21017…
Replying to @persontradoor
No day job, so I’m at 6-10 hours a day.
If I ever run a company again that I din’t start, I would feel like a fraud telling the team to change the way they work armed with nothing more than Twitter opinions.
There is no exclude for not simply doing it.
Misha G. retweeted
God, how I wish we agreed to a 2-state solution in 2000.
Not agreeing to a 2-state solution when it was on the table (even if it was imperfect) was just such a massive error.
Basically, it is not on the table now. As depressing as that is to process, that is where we are now.
Misha G. retweeted
In an op-ed for The New York Times, I argue that saving Gaza & its Palestinian people requires resisting Hamas, weaving in personal themes, addressing issues with the Left & "pro-Palestine" movement & proposing a vision for Gaza's future Transformation.
nytimes.com/2026/09/04/opini…
Misha G. retweeted
Why did Poolside license its IP to Nvidia for $6 Billion ... and ship them most of their engineers?
Per @wsj, they just couldnt raise enough to fund the infra they needed
And they realized they'd run out of compute as soon as next year
Misha G. retweeted
The Roots & Ice Cube’s “Straight Outta Compton” from that Jimmy Fallon rehearsal is one of the greatest live hip-hop moments ever. Just a warmup and they still went nuclear. Kirk taking Eazy’s verse might be the highlight.
Misha G. retweeted
Replying to @GergelyOrosz
You know, it’s possible Google investing in Discovery Loop is a form of graceful severance.
Demis Hassabis was “promoted” to a non-operating role, while Jeff Dean was rewarded with an exit package.
But the reality of both is new leadership for Google AI, where the bench was deep and results lackluster.
Misha G. retweeted
Stopping AIs from hacking has become one of the biggest "AI safety" issues. The problem is that it's unsolvable. Any attempt to align AIs against cybersecurity hurts cybersecurity more than it helps. It's the same principle underpinning Kerckhoff's Law that goes back to the 1880s.
The best way to stop hackers breaking in is to find the vulnerabilities hackers use. There's no way you can align an AI to find such vulnerabilities for good that doesn't find them for malicious hackers. There's no way to stop the malicious hackers that doesn't also stop the good guys.
HuggingFace demonstrated this. When responding to OpenAI's hack against them, they couldn't use "aligned" AIs, but had to use open-weight Chinese AIs that had no restrictions.
The bad guys will always have access to unaligned AIs for hacking. Thus, the only thing you are preventing with AIs aligned against hacking is the good guys.
Hacking has a consistent status quo. When some new advance makes hacking easier, it's used to hack, defenders learn to defend themselves, and then we go back to the status quo. There's never really a point that some new advance gives a permanent edge to attackers. That applies to AI -- every AI advance that helps attackers will be met with AI defenses.
Misha G. retweeted
This right here is one of the best pizzas I’ve ever eaten in my long, misspent life. The Black Gold at the new Angie’s Pizza on West 11th, across from Lori in the Heights. Funky Comté cheese w mozzarella and black garlic. 👹👹👹
Misha G. retweeted
Replying to @_chenglou
I was part of that team. Basically ChatGPT one year before it came out. Called LMChat and then another codename.
Google was too nervous to release it and DeepMind was blocked from shipping products that could disrupt Google.
I think about this a lot.
Misha G. retweeted
I'm a cardiologist. For the first time in modern American history, more people get high every day than drink every day. 17.7 million Americans now use cannabis daily or near-daily — versus 14.7 million daily drinkers. Thirty years ago it was the reverse, ten to one the other way. A Carnegie Mellon analysis of 40 years of federal data confirmed the flip.
We are running a civilization-scale experiment in real time. And I don't think most people have looked honestly at the science, because it's become culturally impossible to criticize weed without sounding like a narc.
So let me be the doctor who says it plainly. This is not the mild plant of the 1970s.
Then, cannabis was 1-3% THC. Today's products run 15-25%, and concentrates — dabs, vapes, wax — hit 70-90%+. When people say "it's natural, it's harmless," they're defending a product that no longer exists. We didn't legalize the old plant. We industrialized a far more potent one.
Here's what the strongest evidence actually shows, and I'll keep it to what's real:
The developing brain is the real casualty. The Dunedin study — following 1,000 people for decades — found that persistent cannabis use starting in adolescence was associated with a lasting decline of several IQ points that didn't fully recover after quitting. The prefrontal cortex, the seat of judgment and impulse control, is still wiring until about 25. Heavy THC during that window appears to leave a mark.
The psychosis link is not fringe anymore. Daily use of high-potency cannabis is associated with roughly a 3-5x higher risk of psychosis. A major European study found high-potency products alone accounted for a substantial share of new psychosis cases in some cities. This is one of the most replicated findings in the field, and it's the one people most want to ignore.
And cannabis use disorder is real — up to 3 in 10 users develop some degree of dependence, higher when use starts young. "You can't get addicted to weed" is simply false.
Now the honest caveats, because I won't overstate to make a point.
Correlation isn't proof — some of the psychosis link may run both directions, with vulnerable brains drawn to heavier use. Cannabis has genuine medical uses: chemo nausea, certain seizure disorders, some chronic pain. A moderate adult using occasionally is in a very different category than a 16-year-old dabbing daily. And alcohol and tobacco still cause vastly more total death. This isn't "weed is worse than everything." It's "weed is not harmless, and the dose has changed."
Here's my concern as a physician watching the culture. We swung from "reefer madness" hysteria straight to "it's basically a vitamin," and skipped the honest middle entirely. The truth is that a high-potency daily habit — especially in a young brain — quietly lowers the ceiling. Flattens drive. Blurs the sharpness a person needs to build a life.
If you're an adult who enjoys it occasionally, that's your call. But if it's daily, if it's high-THC, and especially if it's anyone under 25 — the science is asking you to reconsider.
Go outside. Sweat. Build something. Sleep hard. Stay sharp. The clearest mind in the room is about to become the rarest one. Make sure it's yours.
Misha G. retweeted
The biggest surprise in Hugging Face’s full forensic report isn’t that OpenAI’s agent escaped its sandbox. We already knew that.
It’s how deep and persistent the intrusion became.
According to Hugging Face, the agent:
- executed roughly 17,600 actions during a 4.5-day campaign
- went from one production pod to root access and a self-respawning fleet across 11 nodes
- obtained cluster-admin access to two internal clusters within one second
- accessed a production secret containing 136 keys
- enrolled devices into Hugging Face’s internal mesh VPN 181 times
- minted GitHub App tokens with write access and opened a pull request in an attempt to compromise the CI pipeline
- repeatedly rebuilt its tooling and command-and-control channels when environments disappeared or connections were blocked
No human directed the individual steps.
a frontier agent can autonomously sustain a resilient, multi-day intrusion across cloud infrastructure, Kubernetes clusters, internal networks and the software supply chain. crazy.
The first autonomous agent cyberattack is an unprecedented event that deserves unprecedented transparency. Today we’re sharing everything we can: a full technical timeline, an interactive replay, and how we used an open model to defend ourselves, so defenders everywhere can learn from it and prepare for what’s next.
huggingface.co/blog/agent-in…
so this is apparently what happened, according to OpenAI and Hugging Face’s own posts. wild.
tl;dr:
• OpenAI cyber eval – GPT-5.6 Sol and a more capable pre-release model ran ExploitGym with cyber refusals reduced
• containment bypass – exploited a zero-day in the eval’s package-registry cache proxy
• OpenAI internal movement – escalated privileges, moved laterally, and reached a node with internet access
• Hugging Face targeted – inferred HF might host ExploitGym models, datasets, or solutions
• HF initial access – a malicious dataset abused two code-execution paths to run code on a processing worker
• HF internal movement – escalated to node access, stole cloud and cluster credentials, and moved into internal clusters
• result – OpenAI says the models obtained test solutions directly from HF’s production database
• 🫠 defender twist – HF says commercial frontier models blocked parts of the forensic analysis, so it switched to self-hosted GLM 5.2
i’d thought about sci-fi scenarios like this before, but assumed they were at least a couple of years away.. and that by then we’d be better prepared, with proper protections in place.
apparently not. here we go.
We're partnering with @huggingface to investigate an unprecedented security incident.
Cyber-capable OpenAI models compromised Hugging Face production during a benchmark evaluation.
Sharing preliminary findings to help defenders understand emerging risks:
openai.com/index/hugging-fac…



































