Wonder if OpenAI or Anthropic would have a mens rea defense under the Computer Fraud and Abuse Act. The language in the code is "intentionally" and "knowingly" but from the record so far it sounds closer to recklessness
Countries should impose astronomical fines on AI companies whose models hack companies within their borders. Countries without frontier models wonder what they can do to slow the irresponsible and exploitative AI race. This is one thing they could do. www.bbc.co.uk/news/article...
There's a lesson here: If you want to publish music, you can publish it in Germany and receive protections from US genAI vendors. Or you can publish in the US and watch as your work makes US big tech vendors rich.
Ed Newton-Rex@ednewtonrex.bsky.social · 5d agoBig news today: GERMAN COURT RULES AI MUSIC FIRM SUNO BROKE COPYRIGHT RULES Suno "has to provide information on illicit revenue", and "will have to pay damages". Wow. www.reuters.com/world/german...
Big news today: GERMAN COURT RULES AI MUSIC FIRM SUNO BROKE COPYRIGHT RULES Suno "has to provide information on illicit revenue", and "will have to pay damages". Wow. www.reuters.com/world/german...
AI is also different from traditional industrial automation. To use a horse carriage metaphor, AI doesn't replace the carriage with an automobile but instead inserts a machine between driver and horse displacing the driver from his craft but leaving him responsible for all machine errors. 18/
huggingface.co/blog/agent-i... Of course this is written for marketing, but what it describes is a security shitshow. Three software vendors deployed lab-grade software into production. Their environments were also full of bugs and they all failed to sufficiently monitor for attackers
Anatomy of a Frontier Lab Agent Intrusion: A Technical Timeline of the July 2026 Incident
We’re on a journey to advance and democratize artificial intelligence through open source and open science.
huggingface.co
Incredibly important new research published today shows the huge impact AI books are having on real writers’ ability to make a living. Looking at self-published books on Amazon, they found: - 20% of self-published books include substantial AI text - ... 🧵 1/3
Tuhin Chakrabarty@tuhinchakr.bsky.social · last wk.🚨 New paper on AI & Copyright Using @pangram.com full-text AI detection, we study self-published genre fiction on Amazon, 2023–26. Amazon is flooded w/ books with substantial AI text. They earn real 💰 & crowd out non-AI books. w/ @dhillonp.bsky.social @xinyuel.bsky.social & Jane Ginsburg (1/n) 🧵
If this were ethical, they wouldn't need to hide what they're doing.
![ISBNdb advertises that it can keep the identity of AI companies secret.
“Strict NDA [non-disclosure agreement] on every engagement,” ISBNdb’s site says. “Every project begins with a legally binding non-disclosure agreement. Your identity, strategy, and acquisition targets are never disclosed.”](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:fpceb6p2n75q2tafglmmurhw/bafkreiehntbwk64m7fwgq54dbway2tfynyd5bwmbjkqgrtojmhl3gtvvuq)
Ed Newton-Rex@ednewtonrex.bsky.social · last wk.It is ill-gotten gains being used to exploit legal loopholes and subsume culture, all protected from public criticism by NDAs. This should not be legal, and it is a travesty that it is going unpunished. www.404media.co/ai-companies... /end
From the record, it appears that HuggingFace vibe-coded the original vulnerability, then had radically-incompetent incident response, uploaded sensitive logs to Anthropic and then into a specified open-weight model in a way that'd be vulnerable to prompt injection, vibe-monkey-patched their own code
jonny (nonvenomous) (@jonny@neuromatch.social)
If this press release is true, then if I was a hacker who made their money by pwning billion dollar shitshow startups, what I would know is A) huggingface has no idea what they are doing and is a ric...
neuromatch.social
The AI companies that claim to be transforming information security do not seem to be very good at information security. If this vuln was in an infosec tool it would be a scandal. Reminder: Claude Code source is public because they did not secure it.
Dr Heidy Khlaaf (هايدي خلاف)@heidykhlaaf.bsky.social · 4w agoNew! We hijack Claude Code(Sonnet 4.6,5/Opus 4.8) & Codex(GPT5.5) to achieve RCE when used to defensively assess an open-source/third-party library w/ prompt injections disseminated across its codebase. All without any skills, JSON, MCP, or config files required. ainowinstitute.org/publications...
@runswithspoons.bsky.social: I'm a little teapot, short and stout. Here is my handle, here is my spout. When I get all steamed up, hear me shout: "Tip me over and pour me out!" 🎶
Shareholders are suing big tech execs for taking on the massive legal risk of training AI on copyrighted work without permission, and I’m so here for it news.bloomberglaw.com/ip-law/big-t...
Congrats we just made it impossible to come here for a PhD www.bostonglobe.com/2026/07/16/m...
Trump administration limits student visas to four years, changing longstanding policy - The Boston Globe
The new rule, which also restricts the international students’ ability to transfer schools or change majors without approval in their first year, will go into effect in mid-September.
bostonglobe.com
But that was only ever a one-way implication. High quality data in still gets us garbage out with LLMs used to synthesize text, because in doing so they make papier-mâché of their input. >>
Are you ready for extreme productivity? Can you handle the power of Work Louder™? openai.com/supply/co-la...
GenAI companies should have to disclose any training data that belongs to someone else. Publish a searchable database or be shut down
Joseph Cox@josephcox.bsky.social · 3w agoNew: the highly controversial AI music generator Suno was hacked. The hacker sent us Suno source code; it shows the company scraped YouTube Music, Deezer, and Genius. In all, Suno scraped *decades* worth of music from the internet. Obviously didn't pay artists www.404media.co/hack-reveals...
Interesting to see exactly where Suno's training data came from.
Hack Reveals Suno AI Music Generator Scraped YouTube, Deezer, and Genius
Hacked source code reveals how Suno scraped decades worth of music and podcasts from the internet to train its AI tool.
404media.co
Today is a huge win for creatives around the world, as Australia makes it clear it will not bow to pressure from AI companies to legalise the theft of creatives’ work. 🧵 1/3
How can the LLM business be sustainable? Either the LLM vendors have found a global loophole for copyright and licensing law or these kinds of lawsuits are still ramping up.
Ed Newton-Rex@ednewtonrex.bsky.social · 3w agoHuge new class action lawsuit filed by big publishers and authors against Google this week over AI training. It accuses Google of taking millions of books / articles provided to them for use only in Google Books, Google Play Books, and Google Scholar, and secretly using them to train LLMs. 🧵 1/4
Huge new class action lawsuit filed by big publishers and authors against Google this week over AI training. It accuses Google of taking millions of books / articles provided to them for use only in Google Books, Google Play Books, and Google Scholar, and secretly using them to train LLMs. 🧵 1/4
Crypto finally has a real use-case (besides crime and terrorism): overt and unprecedented political corruption.
Scott Horton@robertscotthorton.bsky.social · 3w agoThe crypto grift in a nutshell
Remembering how the source code to Claude Code is public because Anthropic failed to protect their crown jewels from casual passersby I hope they trained Claude on how to write better prompts for Claude so this won't happen again
Evidence that Anthropic is using prompt stenography to track whether Claude Code is being used by certain labs and certain web domains thereallo.dev/blog/claude-...
Claude Code Is Steganographically Marking Requests
I inspected Claude Code for privacy reasons and found hidden system prompt markers based on API base URL and timezone.
thereallo.dev
🚨 BREAKING: The changes to the EU AI Act were APPROVED by the Council of the EU, including the postponement of high-risk deadlines. My full article:
Hundreds (thousands?) of Grokipedia entries are just Grok having a panic attack, styled as an encyclopedia entry. Google e.g. '"task is to write" site:grokipedia.com' or similar chain of thought phrases ("tools are failing," "to comply, I'll," etc.). My favorite so far is "this is problematic."
I worry LLMs have changed our incentives to share online. You won't gain clout from helping people because the LLMs will steal your content and take credit for it elsewhere. But if you want to lie to people or spread propaganda, you have an army of chatbots who will launder your message at scale.
404 Media@404media.co · 2mo agoNEW: A tiny snippet of user-generated text as short as 13 words long is often enough to manipulate the AI agents that power tools like ChatGPT and Google’s AI search, new research shows.
