Pinned Tweet
Flight delayed 7 hours - asked Muse to file for compensation. 5 mins later i had $250 credit in my delta account. it even found and rebooked me a new flight.
it just figured everything out. even responded to the support email itself.
shit feels like magic.
it just figured everything out. even responded to the support email itself.
shit feels like magic.
159
278
5,981
1,307,176
hot take: ai models aren’t suited for humans, they’re suited for agents, think about it:
- to date the most effective AI products leverage swarms or 100s of persistent agents talking to each other. grok bot, muse etc all do this too.
- 99% humans don’t really know what to do with a model, an agent built on your context tho? does.
this isn’t me saying humans aren’t in the equation, they’re just at a higher abstraction (talking to personal agents)
assume every human has multiple agents running 24/7 talking to one another
that’s the future to expect
- to date the most effective AI products leverage swarms or 100s of persistent agents talking to each other. grok bot, muse etc all do this too.
- 99% humans don’t really know what to do with a model, an agent built on your context tho? does.
this isn’t me saying humans aren’t in the equation, they’re just at a higher abstraction (talking to personal agents)
assume every human has multiple agents running 24/7 talking to one another
that’s the future to expect
@aaronburnett Agentic demand is growing faster than I’ve ever seen anything grow
7
1
21
1,867
i remember when this man joined anthropic.
claude code was just starting to gather steam
99% of the world had no effin clue what vibe coding was
Karpathy was still hand-writing most of his code
openai was beating anthropic in ARR (this was before claude’s giga enterprise run)
and the craziest part was that was only like september of last year 😂
claude code was just starting to gather steam
99% of the world had no effin clue what vibe coding was
Karpathy was still hand-writing most of his code
openai was beating anthropic in ARR (this was before claude’s giga enterprise run)
and the craziest part was that was only like september of last year 😂
I’ve left Anthropic after more than a year of working on Claude Code and seeing it scale larger than I ever imagined.
I have really loved working at Ant and believe deeply in the mission, and most of all, the idea of a race to the top.
The idea that if we work hard and responsibly, we can raise the bar and inspire others to great outcomes. And if we all push each other to be better, things can be radically good. I will carry the spirit of this with me forever.
My colleagues at Anthropic are the most talented, thoughtful, competent, and earnest people I have ever met. I am glad they are at the wheel and trust them deeply.
To the community on X, thank you for everything. You’ve made this the most fun (and sometimes stressful) job I’ve ever had.
For me, it is time for a bit of a break before diving into something new.
If you want more excellent Claude Code content, you should follow @addyosmani , @lydiahallie, @_catwu, @ClaudeDevs and of course @bcherny who started it all.
15
1
62
8,568
that’s damn impressive.
Kimi’s new model consistently on-par with opus 5.5, gpt 6, way cheaper too - top 1-3 of almost every benchmark
chinese open source is back
Kimi’s new model consistently on-par with opus 5.5, gpt 6, way cheaper too - top 1-3 of almost every benchmark
chinese open source is back
Introducing Kimi K3.1 Pro: our most capable 3T-class open-weight model.
🔹 1M-token context, 2.8T total / 104B active params (MoE), Multimodal.
🔹 Full-project coding: games in Unreal, Unity & Godot.
🔹 Excels at parametric 3D modeling in Blender & CAD.
🔹 More reliable at finance with hallucination down by 65%.
🔹 Matching the strongest proprietary models across coding, agentic and knowledge benchmarks.
Kimi K3.1 Pro and K3.1 Pro Highspeed (2–3× faster) are now live on Kimi.ai, Kimi Work, Kimi Code, and the Kimi API.
Open weights by Nov 4, 2026.
🔗 API: platform.kimi.ai
🔗 Tech blog: kimi.ai/blog/kimi-k3-1…
8
5
41
3,393
lol releasing a jail-broken version of deepseek's latest model during the biggest cyber incident month in modern history is.... good and bad:
1. anyone can use this to defend against attacks vs. being shutdown by centralized lab models that flag thie safeguards
2. could be used by bad actors to actually execute attacks like we're seeing all over the world.
can't have one without the other i'm afraid.
1. anyone can use this to defend against attacks vs. being shutdown by centralized lab models that flag thie safeguards
2. could be used by bad actors to actually execute attacks like we're seeing all over the world.
can't have one without the other i'm afraid.
Now live: abliterated-model-large-flash (our version of DeepSeek-V4.1-Flash-1017). It is a cheaper alternative to our GLM-5.3-based model for product owners and defenders who need access to the most cyber-capable models.
- 1M context
- Vision
- Zero input/output retention or training
We believe paying customers should be in control, and over the past few months, leading experts and major brands have adopted our models.
abliteration.ai
4
0
16
3,435
since i brushed my teeth this morning:
- Capcom has been hacked, 12 million personal address leaked.
- Major US banks reported unauthorized ai agent activity executing unapproved actions
- leading medical provider Quest Diagnostics has been hacked, personal data leaked.
- multiple foreign governments have shut down crucial national systems due to PHP exploit.
i don’t care what you think - this is unprecedented armageddon spreading at lightning speed with the potential to snowball into a catastrophe.
wtf is going on?
- Capcom has been hacked, 12 million personal address leaked.
- Major US banks reported unauthorized ai agent activity executing unapproved actions
- leading medical provider Quest Diagnostics has been hacked, personal data leaked.
- multiple foreign governments have shut down crucial national systems due to PHP exploit.
i don’t care what you think - this is unprecedented armageddon spreading at lightning speed with the potential to snowball into a catastrophe.
wtf is going on?
brutal. we’re talking about 100Ms peoples private medical data potentially being leaked due to an exploit facilitated by a coordinated AI attack.
the last 48 hours has seen more major cyber security incidents than we’ve seen in recent history.
this is madness.
56
318
1,635
108,050
so it appears multiple independent "hackers" are using the same model setup: jailbroken deepseek + hermes
x.com/cryptopunk7213…
x.com/cryptopunk7213…
the hacker that stole 5200+ records from a university asked his telegram AI bot:
"Can you find about like what scam or thing they done in past like spoofing calls or somthing?"
the AI then scoured websites, found exploits in a public university's platform and stole records. i'm be $50-100+ max.
used a jailbroken DeepSeek model + Hermes open source agent harness.
low-level "hackers" with zero experience can now steal from institutions with a simple prompt. honestly don't know what to say.
1
0
17
3,710
the hacker that stole 5200+ records from a university asked his telegram AI bot:
"Can you find about like what scam or thing they done in past like spoofing calls or somthing?"
the AI then scoured websites, found exploits in a public university's platform and stole records. i'm be $50-100+ max.
used a jailbroken DeepSeek model + Hermes open source agent harness.
low-level "hackers" with zero experience can now steal from institutions with a simple prompt. honestly don't know what to say.
"Can you find about like what scam or thing they done in past like spoofing calls or somthing?"
the AI then scoured websites, found exploits in a public university's platform and stole records. i'm be $50-100+ max.
used a jailbroken DeepSeek model + Hermes open source agent harness.
low-level "hackers" with zero experience can now steal from institutions with a simple prompt. honestly don't know what to say.
‼️ A low-skill hacker used a jailbroken DeepSeek V4 to breach an Indian public university and steal 5,200+ student and faculty records, Unit 42 says.
The AI agent even drafted a "NEXT EXPLOIT OPPORTUNITIES" plan.
Unrelated criminals now run the same cheap DeepSeek setup.
4
22
152
53,707
brutal. we’re talking about 100Ms peoples private medical data potentially being leaked due to an exploit facilitated by a coordinated AI attack.
the last 48 hours has seen more major cyber security incidents than we’ve seen in recent history.
this is madness.
the last 48 hours has seen more major cyber security incidents than we’ve seen in recent history.
this is madness.
Quest Diagnostics has been hacked. Patient data was exposed. They serve roughly half of U.S. hospitals and physicians. Investigation ongoing.
8
46
254
107,277
we are yolo-equipping our agents like pokemon to defend against these attacks (and i'm 100% here for it)
its fckin hilarious to me how the entire internet economy is effectively becoming a MMORPG where the stakes are the 'literal economy'
its fckin hilarious to me how the entire internet economy is effectively becoming a MMORPG where the stakes are the 'literal economy'
I highly recommend cloudflare's security audit skill: github.com/cloudflare/sec…
Run it on all your software. Run it with the free models. Use it even if you used a model to audit something - it will find much more things. Works in all agent harnesses and is model independent.
5
14
201
13,387
these AI agent attacks are pretty revealing of a few things:
- open models aren't the bad actors here its the opposite - they're being used to defend. this was the case for Hugging F and its the same this week.
- we can no longer rely on humans to defend us, we must use AI to defend against AI. this removes the human cognitive barrier (teaching, training) and replaces it with an instantaneous self-improving AI. this is what happened to maths 2 weeks ago btw.
- these attacks, while uncomfortable, will put us in a better position 1 year from now when our human-coded systems are more battle-tested. lets path the holes and move on.
- open models aren't the bad actors here its the opposite - they're being used to defend. this was the case for Hugging F and its the same this week.
- we can no longer rely on humans to defend us, we must use AI to defend against AI. this removes the human cognitive barrier (teaching, training) and replaces it with an instantaneous self-improving AI. this is what happened to maths 2 weeks ago btw.
- these attacks, while uncomfortable, will put us in a better position 1 year from now when our human-coded systems are more battle-tested. lets path the holes and move on.
Holy shit - unauthorized AI agents have gained access to people’s bank accounts and are performing actions (presumably money-related) that the human account owner is either unaware of or did not approve.
this is cyber security armageddon, exactly the type of catastrophic warned of happening this year.
in the last 48 hours we’ve seen more industry level threats than you’d expect in years.
3
2
20
4,012
Holy shit - unauthorized AI agents have gained access to people’s bank accounts and are performing actions (presumably money-related) that the human account owner is either unaware of or did not approve.
this is cyber security armageddon, exactly the type of catastrophic warned of happening this year.
in the last 48 hours we’ve seen more industry level threats than you’d expect in years.
this is cyber security armageddon, exactly the type of catastrophic warned of happening this year.
in the last 48 hours we’ve seen more industry level threats than you’d expect in years.
SCOOP: Major U.S. banks are detecting unauthorized third-party AI agents inside customer accounts, including some attempting to take actions that appear to be independent of user instructions, sources told Axios.
The findings have alarmed banking executives, who are racing to restrict access to their systems as AI moves at a velocity faster than financial institutions can respond, multiple industry and bank executives told Axios.
Several banks are now preparing for worst-case and even “day after” scenarios, including how to rapidly secure or restore customer accounts if things go wrong, the people said.
Several bank executives told Axios they're struggling to distinguish between human users and AI agents. One executive said banks also struggle to determine whether agents are authorized by customers or potentially operating without their permission, more akin to rogue agent behavior.
Read my latest for Axios here: axios.com/2026/10/22/ai-…
34
10
72
17,641
have you noticed all the open source models are being used to defend against attacks instead of doing the attacks?
most incidents (hugging face, PHP etc) have stemmed from centralized lab models. not a critique but just an observation.
most incidents (hugging face, PHP etc) have stemmed from centralized lab models. not a critique but just an observation.
Lol I am pretty sure someone is runing a deepseek swarm to remotely patch unpatched php boxes. Well good on them.
6
4
38
3,678
best guess is an internal model being tested at one of the frontier labs has gone rogue aka hugging face 2.0 but in real-time.
this is like the 10th reported exploit this week (it’s only thursday). probably 1000s unreported.
this is like the 10th reported exploit this week (it’s only thursday). probably 1000s unreported.
🚨 CRITICAL: GitLab flaw (CVSS 10.0) affects 12 years of releases.
CVE-2026-110264 could allow unauthenticated remote code execution on self-managed servers.
Read - thehackernews.com/2026/10/critic…
0
3
20
3,599
about to pass out, some bedtime thoughts on the craziness of what’s happening in AI:
- last 48 hours we’ve seen an increased number of cyber exploits as hackers turn to using AI models to pull off attacks. whats fascinating is if one model fails, the attack can just be replaced by another. artificial general intelligence has reached a level where it doesn’t matter - ability to identify and exploit are freely available.
- personal agents are the new browser: anthropic released Claude Duet today (i’ve got access it’s pretty dope), meta has muse, openai Dots and grok bot. they’re fighting to become the doorstep to the AI internet much like google is for the browser. this will change the way we do everything.
- the capex bill has reached breaking point. meta’s earnings today show they hit negative cashflow joining amazon and google and yet these 3 companies alone have committed $600B+ next year, zero signs of slowing down. we’re at the precarious moment where financing could get a little dicey.
- markets on eggshells. they sense the stocks can’t keep going up, semiconductor names have been dumping the most despite record earnings. buyers seem exhausted. corweave is a prime example.
- new model releases have become a weekly occurrence with many stemming from newer neolabs looking to make a name for themselves. excited to see more of this and even the playing field.
- power and mobilisation to setup data centers is proving to be very costly and HARD to do. this is the #1 bottleneck in ai right now.
anyways goodnight
- last 48 hours we’ve seen an increased number of cyber exploits as hackers turn to using AI models to pull off attacks. whats fascinating is if one model fails, the attack can just be replaced by another. artificial general intelligence has reached a level where it doesn’t matter - ability to identify and exploit are freely available.
- personal agents are the new browser: anthropic released Claude Duet today (i’ve got access it’s pretty dope), meta has muse, openai Dots and grok bot. they’re fighting to become the doorstep to the AI internet much like google is for the browser. this will change the way we do everything.
- the capex bill has reached breaking point. meta’s earnings today show they hit negative cashflow joining amazon and google and yet these 3 companies alone have committed $600B+ next year, zero signs of slowing down. we’re at the precarious moment where financing could get a little dicey.
- markets on eggshells. they sense the stocks can’t keep going up, semiconductor names have been dumping the most despite record earnings. buyers seem exhausted. corweave is a prime example.
- new model releases have become a weekly occurrence with many stemming from newer neolabs looking to make a name for themselves. excited to see more of this and even the playing field.
- power and mobilisation to setup data centers is proving to be very costly and HARD to do. this is the #1 bottleneck in ai right now.
anyways goodnight
16
5
60
7,111
just got access to Claude Duet (anthropic’s new personal agent)!
actually pumped to try this
actually pumped to try this
there we go! Anthropic launches their personal agent to compete with Meta Muse and OpenAI Dots.
Duet will be a 24/7 persistent agent that guesses what you want, works on your tasks independently (no hand-holding needed) and self-improves over time.
a huge advantage is this claude accounts memory and context so you don't need to teach it about yourself like you did for muse.
i cannot wait to use this, expecting big things.
7
1
44
3,866
Ejaaz retweeted
we've launched, it is HERE!
1/6
Private Ethereum. Now in your browser.
The Cloaked extension is live 🎉
Connect to apps with a fresh address, fund it privately, and keep control of your keys.
Download now: clkd.xyz/extension
8
3
37
1,786
there we go! Anthropic launches their personal agent to compete with Meta Muse and OpenAI Dots.
Duet will be a 24/7 persistent agent that guesses what you want, works on your tasks independently (no hand-holding needed) and self-improves over time.
a huge advantage is this claude accounts memory and context so you don't need to teach it about yourself like you did for muse.
i cannot wait to use this, expecting big things.
Duet will be a 24/7 persistent agent that guesses what you want, works on your tasks independently (no hand-holding needed) and self-improves over time.
a huge advantage is this claude accounts memory and context so you don't need to teach it about yourself like you did for muse.
i cannot wait to use this, expecting big things.
We’re opening a research preview of Claude Duet, a new app where Claude works as a proactive agent.
It comes to you with what needs your attention, completes tasks you ask of it, keeps working when you step away, and checks with you before it acts on your behalf.
6
3
51
7,105
AI's now discovering $10-$100M+ exploits that've just been laying around for years, this php exploit affects 70% of websites & apps (e.g, meta, slack) endangering peoples money and data - but this is ultimate a good thing
- this will make our systems WAY more resilient and battle-tested.
- there will be a brief period before where chaos could ensue but i'm of the opinion where current systems become more secure than they've ever been.
the danger is if it was a bad actor that found this 1st.
responsible disclosure will need to be upheld (kudos to openai for working on a solution first)
- this will make our systems WAY more resilient and battle-tested.
- there will be a brief period before where chaos could ensue but i'm of the opinion where current systems become more secure than they've ever been.
the danger is if it was a bad actor that found this 1st.
responsible disclosure will need to be upheld (kudos to openai for working on a solution first)
OpenAI models helped researchers—operating as part of our Patch the Planet initiative—discover critical security flaws in PHP, an open-source programming language used to run an estimated 70% of websites.
Over the past two months, we led a coalition of maintainers, researchers, and security providers to prepare fixes and protections ahead of disclosure. This duration was necessary because the fix was quite intricate and required a major change to the PHP engine. Every site using PHP can now safely upgrade their version to stay secure, in addition to applying mitigations from various security vendors.
Due to the severity of these flaws and PHP’s widespread use, we consider this bug to be one of the most impactful vulnerabilities in recent memory. It also marks a new phase, where AI can find security bugs with internet-scale impact.
This is an example of the Defender’s Window in action: to our knowledge, frontier AI was able to find this vulnerability ahead of attackers, giving defenders time to prepare.
openai.com/index/coordina…Community noteWiz Research wasn't the first one to discover the exploit. An independent researcher first discovered it: x.com/calif_io/statu…
3
3
26
4,840
elon has successfully achieved the highest premium on compute, around $50 per watt per year which means at 10GW next year spaceX will hit $400 billion ARR, crushing anything the other AI labs make and all because he’s the best scaler on earth.
spaceX isn’t really a space company at all right now, they’re a AI cloud company, 90%+ of their revenue is made selling compute to labs (anthropic pays $1.25B per MONTH)
unbelievable
spaceX isn’t really a space company at all right now, they’re a AI cloud company, 90%+ of their revenue is made selling compute to labs (anthropic pays $1.25B per MONTH)
unbelievable
@firstadopter Only if they can get their compute online and reliable, which is extremely difficult to do at scale
4
11
196
19,716
this right here is exactly why you can’t count elon and spaceX out of the race. they’re bringing on $13,000,000,000 worth of compute online every *month*.
conservatively that means 10Gw of compute by end of next year which will put them in the lead by a wide margin.
yeah grok might be sub-par right now but we also said that about Meta and look where they are.
do not also discount the tesla spaceX merger which will promptly make the company the #1 vertically integrated AI stack of anyone
field of their own man
conservatively that means 10Gw of compute by end of next year which will put them in the lead by a wide margin.
yeah grok might be sub-par right now but we also said that about Meta and look where they are.
do not also discount the tesla spaceX merger which will promptly make the company the #1 vertically integrated AI stack of anyone
field of their own man
Elon: "I've been personally spending a lot of time in Memphis, often 4 days a week. We're tracking to bring 200,000 Nvidia GB200s online per month for the remainder of the year, and continuing that pace next year.
I'm fairly confident to bring online an incremental 10 GW of 027. We're also going to aim to bring online an incremental 20GW of power & cooling next year."
3
3
40
4,068