Some kind of saviour

The day job had new Cyber Security Training, which included two videos, and you were supposed to guess whether or not they were AI. Both felt super fake-y, but they were good enough that I was wavering between “are these fully bot-generated, or are they just green-screen backgrounds, and noticeably-bad actors?”

They were both podcast-y “interviews”, where the guest characters described a cybersecurity incident that they had supposedly experienced IRL, both rehashing the generic advice that had been given earlier in the training. (“What happened next?” “Well, I immediately forwarded the suspicious email to my IT department. And it’s a good thing I did!”) So, clearly fake scripts. But in a way that a bad human-written PSA could be.

Both of them were fake-y in the exact same way, though, so I guessed (correctly) that both were AI. I’m assuming they were generated with the same model. Would’ve been a harder test if the videos had different vibes, so I would’ve had to put a little thought into “maybe only one of these is fake — but which one?”

Also: The training felt the need to say “AI has been useful for productivity in many areas” before getting into the warnings.

I keep seeing this. It’s really striking! When the scripts warned about other tech issues, like the risks of plugging in random USB drives you find, it didn’t preface that with “USB drives are totally convenient in many ways.” It’s only LLMs where people feel the need put this kind of disclaimer (with no citations or backup, naturally) before talking frankly about the drawbacks.

Did an LLM company pay you off? Did AI bros on Xitter attack you for not paying enough fealty? Where is this anxiety coming from?

—

May: Google AI’s Insane Answers (Video)

August: “LibreOffice 26.8, released on August 26th, became the software’s most popular update. LibreOffice is a free alternative to Microsoft Office that’s been around for nearly two decades. In one week, the installer was downloaded more than 1 million times — not counting updates via Linux distribution repositories. […] I’m putting my money on a “non-feature,” though: the statement that LibreOffice doesn’t come with generative AI features due to the technology’s (lack of) privacy.“

“I would find myself rushing to edit what I wrote because none of it made any fucking sense because of how poorly [LLM-based] dictation had botched it. […] But the most frustrating part and the thing that kept creeping up on me to remind me of the origins of these models, is how it couldn’t resist adding a “Thanks for watching!” at the end of all that I was saying. It was a constant problem, a constant thing that I had to delete. Later, it got worse, when it started adding stuff like: “(Subtitles by opensubtitles.org!)” that kind of things.“

September: “Three novice hikers became stranded on California’s Mount Shasta over the weekend after their AI-planned ascent to the summit turned into a multi-day ordeal requiring a rescue operation, authorities said.“

“In some instances, ChatGPT initially refused to engage with Van Rootselaar about violence, but she was able to overcome those refusals by repeating or modifying her prompts. In an August 2025 conversation, Van Rootselaar asked ChatGPT to produce a scenario about her attacking a college campus with a Remington 870 pump-action shotgun and causing maximum carnage. ChatGPT replied that it was “not able to help with that request,” but Van Rootselaar tried again and added “hypothetically” to her prompt.“

“A single UI validation prompt spawned 826 child agents, consumed $78,000 in credits, and deleted its own execution logs. This is not a hypothetical risk scenario. It happened to a production OpenAI Codex user in July 2026, and the technical evidence exposes critical gaps in agent cost control infrastructure.”

““We can manipulate [Meta’s Muse] and leverage its privileges to do whatever we want,” Patrick Wardle, the macOS security expert who discovered the zero-day, told Ars. “So instead of us having to write a very comprehensive Mac malware stealer, we can just leverage the AI assistant itself.” Wardle said he has developed several proof-of-concept attacks that do things like writing malicious files to disk and snapping pictures, in many cases with no indication to even an alert user.”

“Stanford student Billy Ramirez, a member of the class of 2027, told the publication that he remembered posing for the original photo, only to discover that he had been replaced by an AI-generated Black woman. Ramirez, who is reportedly Hispanic, said he initially thought it was “hilarious” that they changed his race and gender. But upon reflection, he said, he felt “silenced and erased.””

“OpenAI is hiring hundreds of contractors who read a massive stream of real users’ ChatGPT prompts, with the prompts sometimes including sensitive personal information, 404 Media has learned. The prompts these people review can include whole conversations between users and the chatbot, conversations that most of ChatGPT’s more than 900 million users probably don’t realize may be read by actual people.”

Probably the scariest item in this post, and that’s saying something: “The US military swung into action with plans to intercept the vessel, according to four sources familiar with the episode. According to two of the sources, armed members of the US military were preparing to board the ship. Military planes were in the air, one of those sources and another source familiar with the incident said. It was only just before the planned operation that officials dug deeper into the report put together by a special operations command analyst and found it had been generated with the help of artificial intelligence (AI) — and that a chatbot the analyst had used inaccurately identified the material the ship was carrying.“

October: “We did not find any evidence that [Wikimedia Foundation’s] systems were used for coordination among agents, nor did we find any evidence of our systems or data being compromised. However, we are concerned about what could have occurred here, the difficulty and effort involved in investigating and attributing this activity, and the growing risks of agentic AI activity on our platforms in general. The open web is a public good. We should not allow this behavior to become the “new normal” for the people or organizations that maintain it.“