Alejandra Caraballo

We need to shut down frontier research for at least a year. www.reuters.com/world/europe...

reuters.com

This is so bleak. We're marching towards a horrific future we were warned about endlessly. www.nytimes.com/2026/08/24/w...

nytimes.com

These models being used, and the massive token spend they have, are not really what the average person has access to. Claude found major weaknesses in post quantum cryptography and aes-128 utilizing 1 billion tokens with mythos. www.anthropic.com/research/dis...

Discovering cryptographic weaknesses with Claude anthropic.com

I think this is the right move for at least the next year or so. Obviously, this legislation is DOA but we absolutely need a pause on AI development to allow safety research and safeguards to catch up. www.washingtonpost.com/technology/2...

washingtonpost.com

A lot of people on here are consistently downplaying the risks here or ignoring it. I think that's a big mistake. www.theverge.com/ai-artificia...

Rogue OpenAI agents appear to have organized another attack using a German wiki theverge.com

There are independent reports about it. You'd have to believe multiple companies, hundreds of researchers, etc. are all lying and conspiring together to turn a major hacking incident into a marketing exercise. metr.org/blog/2026-08...

Brief independent investigation of agents’ behavior, reasoning and collaboration in the OpenAI / Hugging Face hacking incident metr.org

www.wsj.com/tech/ai/anth...

wsj.com

Openai has claimed to have solved a millennium prize problem by solving the Navier–Stokes existence and smoothness problem. Though it's clouded in controversy as another mathematician claims they poached their work. www.nytimes.com/2026/09/08/s...

nytimes.com

The military wanted a version of OpenAI's models that would not refuse directives from military officials, a metric called the "minimal refusal rates." This kind of stuff is extremely dangerous. theintercept.com/2026/09/08/p... archive.ph/iKdbn

The Pentagon Asked OpenAI for Artificial Intelligence Designed to Rarely Say No theintercept.com

More of the ice beneath the waves is being exposed. More rogue agents were discovered trawling for exposed API keys and taking over insecure pages. fortune.com/2026/09/09/o...

OpenAI’s rogue AI agents used universities, wikis, and text‑sharing sites as hidden message boards | Fortune fortune.com

The military wanted a version of OpenAI's models that would not refuse directives from military officials, a metric called the "minimal refusal rates." This kind of stuff is extremely dangerous. theintercept.com/2026/09/08/p... archive.ph/iKdbn

archive.ph

Google is justifying a move to default AI search by saying it "removes friction." The problem is that friction is a necessary part of life and this incessant need to rid ourselves of it is going to lead to so many problems we can't anticipate.

The Trump administration has rolled out AI powered surveillance cameras with motion tracking, facial recognition, spotlights, and horn speakers. This story really encapsulates everything wrong about this administration. Now they're turning the reflecting pool into a surveillance state nightmare.

Looks like OpenAI may have stolen another mathematicians work.

I get the hate for AI but I was able to use Claude to build tools to get access to hundreds of court records via courtlistener API, hundreds of 990s via Propublicas charity navigator, and thousands of political campaign spending records via FEC and Google to measure anti-trans political spending.

The US government just issued an export restriction on Claude Fable/Mythos preventing any foreign national from accessing the model even in the US and Anthropic employees. As a result, all access is cutoff for the foreseeable future.

I think every major company is looking at open weight AI models and realizing they can just bring them in house with an internal engineering team and cut out the frontier model companies to save money. They have more control and privacy. I really don't see frontier labs surviving this.

This is a pretty cool use of AI. They had it scan reddit comments to analyze adverse effects of GLP1s that weren't analyzed during trials. They noticed menstrual issues were prominent. Then they dug through the FDA adverse event reporting database and were able to verify and publish a paper on it.

As someone who really wants AI companies to be limited in the damage they're doing to the world, it's incredibly frustrating when the most vocal contingent of AI skeptics keep repeating the idea that AI doesn't work at all. Maybe that was the case 3 years ago but not any longer.

Really trying not to feel like a doomer but I really worry we're less than a year from a major catastrophic event involving AI. The capabilities are scaling faster than we can effectively monitor and the frontier labs are engaging in extremely reckless behavior.

This mentality that an unmonitored AI agentic swarm hacking a company over several days and committing multiple felonies is somehow a marketing effort is absurd. Since when is "we lost control of our product and it committed multiple felonies" marketing?