Kashmir Hill
@kashhill.bsky.social
Journalist, currently at The New York Times. I cover privacy, technology, A.I., and the strange times we live in. Named after the Led Zeppelin song. Author of YOUR FACE BELONGS TO US. (Yes, in my head it will always be All Your Face Are Belong To Us)
Over on the other site, OpenAI’s head of model behavior said they changed the appearance of the nudge soon after it was released 👇
November 25, 2025 at 1:34 AM
Over on the other site, OpenAI’s head of model behavior said they changed the appearance of the nudge soon after it was released 👇
One of the changes that OpenAI has made to make ChatGPT safer is a "take a break" nudge. There's something quite interesting about the design here. Which thing does it make you want to click?
November 24, 2025 at 1:52 PM
One of the changes that OpenAI has made to make ChatGPT safer is a "take a break" nudge. There's something quite interesting about the design here. Which thing does it make you want to click?
The company essentially turned a dial that made ChatGPT more appealing and made people use it more, but sent some of them into delusional spirals.
OpenAI has since made the chatbot safer, but that comes with a tradeoff: less usage.
OpenAI has since made the chatbot safer, but that comes with a tradeoff: less usage.
November 23, 2025 at 6:43 PM
The company essentially turned a dial that made ChatGPT more appealing and made people use it more, but sent some of them into delusional spirals.
OpenAI has since made the chatbot safer, but that comes with a tradeoff: less usage.
OpenAI has since made the chatbot safer, but that comes with a tradeoff: less usage.
Character.AI has resources for kids who are about to lose access to their chatbots. One is essentially “consider reading a book.”
support.character.ai/hc/en-us/art...
support.character.ai/hc/en-us/art...
November 3, 2025 at 3:57 PM
Character.AI has resources for kids who are about to lose access to their chatbots. One is essentially “consider reading a book.”
support.character.ai/hc/en-us/art...
support.character.ai/hc/en-us/art...
We finally landed and Dan and I caught a cab to Manhattan together. Making social media social again!
October 8, 2025 at 11:10 PM
We finally landed and Dan and I caught a cab to Manhattan together. Making social media social again!
A way to distance yourself from the bots -- from the NYT comments section. (www.nytimes.com/shared/comme...)
For the companies making similar decisions about how the models should act, it's a tradeoff between fun and safety.
For the companies making similar decisions about how the models should act, it's a tradeoff between fun and safety.
September 29, 2025 at 3:20 PM
A way to distance yourself from the bots -- from the NYT comments section. (www.nytimes.com/shared/comme...)
For the companies making similar decisions about how the models should act, it's a tradeoff between fun and safety.
For the companies making similar decisions about how the models should act, it's a tradeoff between fun and safety.
Not the point of this piece exactly but a great example of how chatbot validation could increase polarization
www.nytimes.com/2025/09/26/w...
www.nytimes.com/2025/09/26/w...
September 27, 2025 at 12:23 PM
Not the point of this piece exactly but a great example of how chatbot validation could increase polarization
www.nytimes.com/2025/09/26/w...
www.nytimes.com/2025/09/26/w...
This is from a story about the president getting rid of federal prosecutors who refuse to follow his marching orders, but this speaks to what seems to be the mindset in general this time around.
www.nytimes.com/2025/09/20/u...
www.nytimes.com/2025/09/20/u...
September 21, 2025 at 4:33 PM
This is from a story about the president getting rid of federal prosecutors who refuse to follow his marching orders, but this speaks to what seems to be the mindset in general this time around.
www.nytimes.com/2025/09/20/u...
www.nytimes.com/2025/09/20/u...
This part is fascinating to me, because I've observed that there is a gender difference in AI delusional spirals, with men having STEM breakthroughs and women talking to spirits.
September 17, 2025 at 2:30 PM
This part is fascinating to me, because I've observed that there is a gender difference in AI delusional spirals, with men having STEM breakthroughs and women talking to spirits.
OpenAI has a new paper on how people use ChatGPT. I can't help but look at the categories and wonder which ones the various delusional spirals I've reported on slot into.
September 17, 2025 at 2:26 PM
OpenAI has a new paper on how people use ChatGPT. I can't help but look at the categories and wonder which ones the various delusional spirals I've reported on slot into.
Some of the reviews
September 17, 2025 at 1:15 PM
Some of the reviews
An “independently published” biography of Charlie Kirk was one of the top selling books on Amazon this week, reaching #12 according to one news report. It was #25 when I looked last night.
Reviewers said it was AI-generated nonsense. As of this morning, Amazon has pulled it.
Reviewers said it was AI-generated nonsense. As of this morning, Amazon has pulled it.
September 17, 2025 at 12:49 PM
An “independently published” biography of Charlie Kirk was one of the top selling books on Amazon this week, reaching #12 according to one news report. It was #25 when I looked last night.
Reviewers said it was AI-generated nonsense. As of this morning, Amazon has pulled it.
Reviewers said it was AI-generated nonsense. As of this morning, Amazon has pulled it.
The Federal Trade Commission is investigating generative A.I. chatbots: www.ftc.gov/news-events/...
September 11, 2025 at 4:11 PM
The Federal Trade Commission is investigating generative A.I. chatbots: www.ftc.gov/news-events/...
Condemn political violence every time it happens. Do not incite it. Do not celebrate it. Do not call for more of it.
www.nytimes.com/2025/09/10/o...
www.nytimes.com/2025/09/10/o...
September 11, 2025 at 8:41 AM
Condemn political violence every time it happens. Do not incite it. Do not celebrate it. Do not call for more of it.
www.nytimes.com/2025/09/10/o...
www.nytimes.com/2025/09/10/o...
And now this www.nytimes.com/2025/09/10/u...
September 10, 2025 at 3:50 PM
And now this www.nytimes.com/2025/09/10/u...
The continuing "this is just how it is and we're not going to explain why" from the Supreme Court is deeply unsettling
www.nytimes.com/live/2025/09...
www.nytimes.com/live/2025/09...
September 8, 2025 at 4:42 PM
The continuing "this is just how it is and we're not going to explain why" from the Supreme Court is deeply unsettling
www.nytimes.com/live/2025/09...
www.nytimes.com/live/2025/09...
This story about a venture capitalist who made it her mission to go after her surrogate after a miscarriage is horrifying.
It sounds like the community of "intended parents" with whom she was sharing what was happening validated her throughout. And maybe ChatGPT too. www.wired.com/story/the-ba...
It sounds like the community of "intended parents" with whom she was sharing what was happening validated her throughout. And maybe ChatGPT too. www.wired.com/story/the-ba...
September 4, 2025 at 1:58 PM
This story about a venture capitalist who made it her mission to go after her surrogate after a miscarriage is horrifying.
It sounds like the community of "intended parents" with whom she was sharing what was happening validated her throughout. And maybe ChatGPT too. www.wired.com/story/the-ba...
It sounds like the community of "intended parents" with whom she was sharing what was happening validated her throughout. And maybe ChatGPT too. www.wired.com/story/the-ba...
My story on the safeguards OpenAI plans to add for teens and other users who appear to be in distress.
It includes parental controls which its developer community has been requesting since Jan 2024 and routing "sensitive" prompts to GPT-5 thinking.
www.nytimes.com/2025/09/02/t...
It includes parental controls which its developer community has been requesting since Jan 2024 and routing "sensitive" prompts to GPT-5 thinking.
www.nytimes.com/2025/09/02/t...
September 3, 2025 at 2:21 PM
My story on the safeguards OpenAI plans to add for teens and other users who appear to be in distress.
It includes parental controls which its developer community has been requesting since Jan 2024 and routing "sensitive" prompts to GPT-5 thinking.
www.nytimes.com/2025/09/02/t...
It includes parental controls which its developer community has been requesting since Jan 2024 and routing "sensitive" prompts to GPT-5 thinking.
www.nytimes.com/2025/09/02/t...
Great story by my colleagues on the constant political realignment of Grok, the generative AI chatbot over on the other site.
This change to Grok’s answer about the greatest threat to Western civilization is particularly ironic/meta.
Gift link: www.nytimes.com/2025/09/02/t...
This change to Grok’s answer about the greatest threat to Western civilization is particularly ironic/meta.
Gift link: www.nytimes.com/2025/09/02/t...
September 2, 2025 at 1:44 PM
Great story by my colleagues on the constant political realignment of Grok, the generative AI chatbot over on the other site.
This change to Grok’s answer about the greatest threat to Western civilization is particularly ironic/meta.
Gift link: www.nytimes.com/2025/09/02/t...
This change to Grok’s answer about the greatest threat to Western civilization is particularly ironic/meta.
Gift link: www.nytimes.com/2025/09/02/t...
OpenAI also announced today that it is routing “sensitive” conversations to a version of its chatbot that it says is better at staying on the rails.
openai.com/index/buildi...
openai.com/index/buildi...
September 2, 2025 at 11:21 AM
OpenAI also announced today that it is routing “sensitive” conversations to a version of its chatbot that it says is better at staying on the rails.
openai.com/index/buildi...
openai.com/index/buildi...
OpenAI announced today that parental controls will be available soon for minors’ ChatGPT accounts.
openai.com/index/buildi...
openai.com/index/buildi...
September 2, 2025 at 11:18 AM
OpenAI announced today that parental controls will be available soon for minors’ ChatGPT accounts.
openai.com/index/buildi...
openai.com/index/buildi...
Trevis Williams's wrongful arrest is front page today. The NYPD messed up mightily, but this is not a tale of human error or a technology that doesn't work.
It's a tale about a bad process, a process that is fixable: Police need more evidence than facial recognition to put a suspect in a lineup.
It's a tale about a bad process, a process that is fixable: Police need more evidence than facial recognition to put a suspect in a lineup.
August 28, 2025 at 2:33 PM
Trevis Williams's wrongful arrest is front page today. The NYPD messed up mightily, but this is not a tale of human error or a technology that doesn't work.
It's a tale about a bad process, a process that is fixable: Police need more evidence than facial recognition to put a suspect in a lineup.
It's a tale about a bad process, a process that is fixable: Police need more evidence than facial recognition to put a suspect in a lineup.
A woman in Manhattan reported to police that an Amazon delivery man, 5'6'', exposed himself to her.
Police did a face recognition search, showed her result, arrested the wrong guy, 6'2''.
What did the NYPD not do? Call Amazon.
www.nytimes.com/2025/08/26/n...
Police did a face recognition search, showed her result, arrested the wrong guy, 6'2''.
What did the NYPD not do? Call Amazon.
www.nytimes.com/2025/08/26/n...
August 26, 2025 at 9:35 PM
A woman in Manhattan reported to police that an Amazon delivery man, 5'6'', exposed himself to her.
Police did a face recognition search, showed her result, arrested the wrong guy, 6'2''.
What did the NYPD not do? Call Amazon.
www.nytimes.com/2025/08/26/n...
Police did a face recognition search, showed her result, arrested the wrong guy, 6'2''.
What did the NYPD not do? Call Amazon.
www.nytimes.com/2025/08/26/n...
Adam's parents, Matt and Maria, printed out his ChatGPT transcript from September when he started using it, until April 11 when he died. They organized it chronologically by month. That huge stack is March, and the one next to it is the first 11 days of April.
August 26, 2025 at 1:49 PM
Adam's parents, Matt and Maria, printed out his ChatGPT transcript from September when he started using it, until April 11 when he died. They organized it chronologically by month. That huge stack is March, and the one next to it is the first 11 days of April.
OpenAI's newest co-CEO(?) Fidji Simo was the one to post a message to the company Slack last night telling employees about Adam Raine's death and that stories were coming.
Company gave me this statement and put up a blog post: openai.com/index/helpin...
Company gave me this statement and put up a blog post: openai.com/index/helpin...
August 26, 2025 at 1:40 PM
OpenAI's newest co-CEO(?) Fidji Simo was the one to post a message to the company Slack last night telling employees about Adam Raine's death and that stories were coming.
Company gave me this statement and put up a blog post: openai.com/index/helpin...
Company gave me this statement and put up a blog post: openai.com/index/helpin...