Anthropic: "We're profitable on an adjusted basis!"

Also Anthropic: "We can't continue to operate without a $42 billion loan from one of our chip suppliers"

https://www.reuters.com/business/broadcom-lend-anthropic-up-42-billion-lease-its-chips-filing-says-2026-10-01/

Jesus H. Christ.

#ai

Shocker: The #ai doomer whistleblower Jacob Coxon got help from a PR firm to do his press tour.

And it's the same firm that represents effective altruist and sex cult leader Eliezer "Yud" Yudkowsky.

Tell me how a 27 year old hires a top tier PR firm on a whistleblower story?

https://tech.yahoo.com/ai/meta-ai/articles/ex-anthropic-researcher-warned-ai-082244257.html

P.S. The article mistakenly says Timnit Gebru was represented by DEY (source: https://x.com/timnitGebru/status/2103532550958592038/photo/1)

"It is easier for us to imagine the Terminator than it is for us to imagine Sam Altman going to jail"

This captures #ai problem precisely. It's not software. It's laziness.

It's the intellectual dishonesty of solopsistic billionaires who don't understand what they're buying.

It's the laziness of the media who simply take all claims at face value.

It is the laziness of our politicians who legislate around ghost stories than enforce the laws already on the books.

https://www.youtube.com/watch?v=t3UlAQuVbmE

We are literally living out a scene from Parks & Recreation. Except instead of Zorp, it's AI that will kill everyone. And the group is called the Rationalists.

Don't believe me? Check out their wiki: https://en.wikipedia.org/wiki/Rationalist_community

Please stop telling me that AI has gone rogue. It hasn't. We're simply being sold a bridge by the CEOs of glorified R&D labs that should be put on trial for violating the Computer Fraud & Abuse Act.

#ai

"The reasoning model gooses revenue. The orchestration layer kills revenue."

"The AI stack is 90% code and 10% model."

Eli The Computer Guy reads my mind.

https://www.youtube.com/watch?v=sp__3whhATU

#ai

Haven't given an update on my zero-trust agent framework Sycophant in a while. But it now natively supports in-cluster inference.

So if you have two computers and only one of them is powerful enough to host a model you can do that.

I'll playing around with some CPU-optimized LLMs to figure out reliable model routing once I finish cleaning up.

https://github.com/calebfaruki/sycophant

#ai #openclaw #agenticai

Between cookie banners and claude watermarking, does the EU ever ask itself if it actually understands what they fuck they're doing?

So much mental effort exhausted to navigate dark patterns and word vomit.

#ai #gdpr

Anna's Archive 🤙🤙🤙

https://www.tomshardware.com/tech-industry/artificial-intelligence/worlds-largest-open-library-calls-for-volunteers-to-scan-and-preserve-physical-books-as-ai-companies-buy-scan-and-destroy-them-annas-archive-says-time-is-running-out-as-knowledge-is-permanently-monopolized-on-private-servers

Excited about Count Binface? Wait until you hear about Binjamin Netanyahu

If I was a betting man, I would certainly say that virtually every major tech company owns an Israeli startup that operates with complete autonomy for the purpose of covert government operations.

You never hear about these acquisitions. You never hear about their value. But acquisitions are almost always taken as a positive stock price signal. Why do we not hear about these?

This AI-gone-rogue narrative is just a total smoke screen.

https://www.cnbc.com/2026/08/09/israeli-startup-irregular-linked-to-ai-hacks-openai-anthropic-meta.html

One of the most controversial takes of our generation. This will certainly upset the slow poopers.

https://nate.spot/no-phone-while-poop/

Even running an #llm locally won't save you

Open weights give you the numbers, not the meaning, so you can't prove it isn't backdoored.

Poisoning is cheap. ~250 documents is enough to backdoor a model, no matter how big. And the place you download from, Hugging Face, just got breached.

Models are adversarial. That's the premise Sycophant is built on. Zero trust architecture. Defense in depth. Mechanically verifiable security invariants.

https://github.com/calebfaruki/sycophant

#ai #infosec #rust #kubernetes

The tragic irony of having a #fairphone is that you can definitely repair your phone yourself. But no one sells the parts locally.

So even if you believe in #righttorepair , you certainly can't exercise that right when the fix is urgent.

Repair will never supersede replacement if it takes you a week longer to do the right thing.

I built Airlock last week around a simple idea: agent containers shouldn’t have your CLI credentials. It worked. But proxying the LLM runtime showed that a shim isn’t enough. You also need user input, MCP tools, and conversation history. So I built Tightbeam: keep the LLM client and credentials on the host, stream output + tool calls into the container over a Unix socket, and execute only in the workspace.

The container is no longer an agent.

https://github.com/calebfaruki/tightbeam

#ai #llm #docker #infoSec

Airlock v0.3.0: command modules are now opt-in.

Airlock already shipped hardened deny rules per tool and scoped each container via profiles. Now there's a third layer: no command loads unless the operator enables it.

SSH is worth calling out. It's remote code execution with real keys. If you enable it, 13 deny rules lock it to single-hop, one-off commands. No tunneling, no forwarding, no credential passthrough.

https://github.com/calebfaruki/airlock

#infosec #docker #opensourcesecurity #aiagents #owasp #ai

If you mount SSH keys into Docker for an AI coding agent, that agent can do whatever it wants with them. Push to any repo. Access any server. Frameworks that remove HITL mean nobody's watching.

Airlock v0.2.0 fixes this. Containers never hold credentials. A host-side daemon proxies CLI tools over a unix socket. Each container gets a profile with only the commands and credentials it needs. Nothing more.

https://github.com/calebfaruki/airlock/releases/tag/v0.2.0

#InfoSec #Docker #OpenSource #DevSecOps #AI #Rust

We've spent a lot of time since we launched on International Repair Day to bring tons of new features and improvements to @therepairapp. I hope that your #newyearsresolution will be to try and repair one thing.

Please DM me if you end up using The Repair App. All feedback is welcome.

It's forever free. If there's a feature you want, let me know.

We've got a few more announcements in the coming days as well! So we hope you'll stay tuned.

#righttorepair #repaircafe #repair

I'm convinced that the #rust hype train is a means to wrest control of business critical apps from the #opensource community to proprietary source startups.

Software is a liability not an asset. Don't fall for the hype. Trust in your experience.

It's only been a few days since we launched The Repair App. The initial feedback we've received has been overwhelmingly supportive. Though lots of work remains.

The most interesting item needing repair was a a Nakamichi Dragon. A hi-fi cassette deck! Finding repair shops that can fix this kind of thing is definitely an adventure.

#righttorepair #audiophile