It's like the olympics of intellectual property law out here, and it's hard to keep up. Not to mention, the US economy has been swallowed whole by the tech sector, who has in turn managed to convince everyone that 1. all your base are belong to us, and 2. resistance is futile.
They created a robot that steals the output we use as evidence of human agency (art, literature, etc.), and uses that to pretend it's a human (therapy bot, companion bot, etc.), and then it lies to, cheats on, and steals from other people, just like a real person! At this point it's hard to tell the difference, and soon, it won't matter.
Here's a taste of how crazy things have gotten - it's possible that if you let an AI-agent (aigent?) take over your computer and use your browser, then you may be attacked by a person who doped online forum (reddit) posts with invisible but malicious prompt instructions to reset email passwords.
Researchers suggest OpenAI trained AI models on paywalled O'Reilly books
Apr 2025, TechCrunch
It isn’t a smoking gun, the co-authors are careful to note. They acknowledge that their experimental method isn’t foolproof and that OpenAI might’ve collected the paywalled book excerpts from users copying and pasting it into ChatGPT.
Totally unrelated above image credit: AI Art - Uterus 1 - 2025
Prominent chatbots routinely exaggerate science findings, study shows
May 2025, phys.org
When summarizing scientific studies, large language models like ChatGPT and DeepSeek produce inaccurate conclusions in up to 73% of cases.Surprisingly, prompts for accuracy increased the problem and newer LLMs performed worse than older ones.Six of ten models systematically exaggerated claims found in the original texts, often in subtle but impactful ways; for instance, changing cautious, past-tense claims like "The treatment was effective in this study" to a more sweeping, present-tense version like "The treatment is effective." These changes can mislead readers into believing that findings apply much more broadly than they actually do.Strikingly, when the models were explicitly prompted to avoid inaccuracies, they were nearly twice as likely to produce overgeneralized conclusions than when given a simple summary request."This effect is concerning"
via Utrecht University: Uwe Peters et al, Generalization bias in large language model summarization of scientific research, Royal Society Open Science (2025). DOI: 10.1098/rsos.241776
Anthropic's new AI model turns to blackmail when engineers try to take it offline
May 2025, TechCrunch
During pre-release testing, Anthropic asked Claude Opus 4 to act as an assistant for a fictional company and consider the long-term consequences of its actions. Safety testers then gave Claude Opus 4 access to fictional company emails implying the AI model would soon be replaced by another system, and that the engineer behind the change was cheating on their spouse.This was found a while back as well: "When asked if it used the insider information, the bot denies it" (BBC News 2023).
via Anthropic’s Claude Opus 4 model release: https://www-cdn.anthropic.com/4263b940cabb546aa0e3283f35b686f4f3b2ff47.pdf
It turns out you can train AI models without copyrighted material
Jun 2025, Engadget
Ethically-Sourced Dataset - you know like 130,000 books in the Library of Congress
Here's the paper: https://github.com/r-three/common-pile/blob/main/paper.pdf
Why Was Nvidia Hosting Blogs About 'Brazilian Facesitting Fart Games'?
Jun 2025, 404 Media
Long story short, somebody found an abandoned Nvidia website, and filled it with AI-generated websites, all with crazy ads. Same with vaccines.gov, the American Council on Education, Stanford, and NPR, all on subdomains in varying states of useability.
But this summary comes from user msmash on slashdot, "The operation exploits search engines' trust in institutional domains, with Google's AI Overview already serving the fabricated content as factual information to users searching for local businesses."
Maine police department apologizes for AI-altered evidence photo
Jul 2025, phys.org
"AI-enhanced" should mean "unusable", but in this case, they're just trying to do really basic photo editing, cropping a photo, and then all of the sudden it changes that actual contents of the photo.
The image from the Police Department showed a collection of drug paraphernalia purportedly seized during a recent drug bust, including a scale and white powder in plastic bags. According to police, an officer involved in the arrests snapped the evidence photo and used a photo editing app to insert the department’s patch.“It was never our intent to alter the image of the evidence. We never realized that using a photoshop app to add our logo would alter a photograph so substantially.”
"photoshop app"
Topological approach detects adversarial attacks in multimodal AI systems
Aug 2025, phys.org
I do not recall having seen the word topological in the context of AI, but this is about multimodal:
When an attack disrupts the geometric alignment of text and image embeddings, it creates a measurable distortion. The researchers developed two pioneering techniques, dubbed "topological-contrastive losses," to quantify these topological differences with precision, effectively pinpointing the presence of adversarial inputs.
via Los Alamos National Laboratory: Minh Vu et al, Topological Signatures of Adversaries in Multimodal Alignments, arXiv (2025). DOI: 10.48550/arxiv.2501.18006
Is AI really trying to escape human control and blackmail people?
Aug 2025, Ars Technica
On agency; let's clear this up with a good explanation by Ars writer Benj Edwards:
Consider a self-propelled lawnmower that follows its programming: If it fails to detect an obstacle and runs over someone's foot, we don't say the lawnmower "decided" to cause injury or "refused" to stop. We recognize it as faulty engineering or defective sensors. The same principle applies to AI models - which are software tools - but their internal complexity and use of language make it tempting to assign human-like intentions where none actually exist.In a way, AI models launder human responsibility and human agency through their complexity. When outputs emerge from layers of neural networks processing billions of parameters, researchers can claim they're investigating a mysterious "black box" as if it were an alien entity.But the truth is simpler: These systems take inputs and process them through statistical tendencies derived from training data. The seeming randomness in their outputs - which makes each response slightly different - creates an illusion of unpredictability that resembles agency. Yet underneath, it's still deterministic software following mathematical operations. No consciousness required, just complex engineering that makes it easy to forget humans built every part of it. (Thanks Benj)
Anthropic’s auto-clicking AI Chrome extension raises browser-hijacking concerns
Aug 2025, Ars Technica
Last week, Brave's security team discovered that Perplexity's Comet browser could be tricked into accessing users' Gmail accounts and triggering password recovery flows through malicious instructions hidden in Reddit posts. When users asked Comet to summarize a Reddit thread, attackers could embed invisible commands that instructed the AI to open Gmail in another tab, extract the user's email address, and perform unauthorized actions. Although Perplexity attempted to fix the vulnerability, Brave later confirmed that its mitigations were defeated and the security hole remained.
AI startup Anthropic agrees to pay $1.5bn to settle book piracy lawsuit
Sep 2025, The Guardian
The artificial intelligence company Anthropic has agreed to pay $1.5bn to settle a class-action lawsuit by book authors who say the company took pirated copies of their works to train its chatbot.The company has agreed to pay authors about $3,000 for each of an estimated 500,000 books covered by the settlement.“As best as we can tell, it’s the largest copyright recovery ever. It is the first of its kind in the AI era.”
(Congrats to the Author's Guild on this one, and note they have a Human Authored Certification system for all guild members to prove they used no AI in their work.) https://authorsguild.org/human-authored/
AI chatbots routinely use user conversations for training, raising privacy concerns
Oct 2025, phys.org
Is this the second or third law of data collection? (answer below!)
In the case of multiproduct companies, such as Google, Meta, Microsoft, and Amazon, user interactions also routinely get merged with information gleaned from other products consumers use on those platforms—search queries, sales/purchases, social media engagement, and the like.
In other words, your public comments on the neighborhood surveillance app, your face from the biodata verification app, the items left in your online shopping cart, the locations you visit taken from the municipal FLOC network, and every query you ever typed into a searchbox, are all getting smashed together.
via Stanford: Jennifer King et al, User Privacy and Large Language Models: An Analysis of Frontier Developers' Privacy Policies, arXiv (2025). DOI: 10.48550/arxiv.2509.05382
The 4 Laws of Data Dynamics
1. Data must seek and merge with complementary data.
2. Data always will be used for purposes other than originally intended.
3. Data collected about individuals will be used to cause harm.
4. Confidential information is confidential only until someone decides it's not. (p14)
--The Naked Consumer: How Our Lives Become Public Commodities, Erik Larson, Henry Holt Publishers, 1992 (read the full post: https://networkaddress.blogspot.com/2022/05/the-naked-consumer.html)
So the incident above would be an example of both the first and second laws of data collection. The third law, harm, comes next (because that's what happens after the bubble pops, of course).
