Ryan J. Gallagher
@ryanjgallag.com
Applied scientist trying to make the internet a little better. PhD. Trust & safety, platform manipulation, networks, fingerstyle guitar. I use my hair to express myself. They/he
Pinned
Ryan J. Gallagher
@ryanjgallag.com
· Nov 12
With more people migrating to Bluesky (and my general quietness since Twitter fell apart), I want to reintroduce myself!
I'm an applied scientist in online trust & safety. I specialize in detecting coordinated/harmful/deceptive networks, and bridging domain expertise with machine learning systems
I'm an applied scientist in online trust & safety. I specialize in detecting coordinated/harmful/deceptive networks, and bridging domain expertise with machine learning systems
Reposted by Ryan J. Gallagher
Two days ago it was Elissa Slotkin. Last week it was Jerome Powell. Before that, Mark Kelly. Weaponizing the justice system against your opponents is an authoritarian tactic.
The only person not being investigated for the shooting of Renee Good is the federal agent who shot her.
The only person not being investigated for the shooting of Renee Good is the federal agent who shot her.
Breaking news: The Justice Department has issued subpoenas for Minnesota Gov. Tim Walz and Minneapolis Mayor Jacob Frey as part of an investigation alleging that they are impeding federal law enforcement officers’ abilities to do their jobs in the state.
Justice Dept. launches criminal investigation of Minnesota governor
The Justice Department has issued subpoenas for Minnesota Gov. Tim Walz and Minneapolis Mayor Jacob Frey as part of an investigation into impeding law enforcement.
www.washingtonpost.com
January 16, 2026 at 11:46 PM
Two days ago it was Elissa Slotkin. Last week it was Jerome Powell. Before that, Mark Kelly. Weaponizing the justice system against your opponents is an authoritarian tactic.
The only person not being investigated for the shooting of Renee Good is the federal agent who shot her.
The only person not being investigated for the shooting of Renee Good is the federal agent who shot her.
Reposted by Ryan J. Gallagher
ICE has kidnapped so many people from their cars that the city had to put out this statement. Let that sink in.
Vehicles that are abandoned due to an ICE detention and towed to the City's impound lot will be released to their owners or a representative at no cost.
Learn more: https://www.minneapolismn.gov/getting-around/parking-driving/impound-lot/#d.en.200746
Learn more: https://www.minneapolismn.gov/getting-around/parking-driving/impound-lot/#d.en.200746
January 16, 2026 at 11:42 PM
ICE has kidnapped so many people from their cars that the city had to put out this statement. Let that sink in.
Reposted by Ryan J. Gallagher
#OutNow in #iCS
From reading and sharing to curating and disconnecting, this paper explains how safety structures political participation inside everyday messaging groups.
doi.org/10.1080/1369...
From reading and sharing to curating and disconnecting, this paper explains how safety structures political participation inside everyday messaging groups.
doi.org/10.1080/1369...
From connective to disconnective action: how safety shapes political participation repertoires in instant messaging groups
Safety is a fundamental yet increasingly precarious condition for political participation in an era of democratic backsliding and a hyper-visible digital environment. Often overlooked in political ...
doi.org
January 16, 2026 at 10:36 AM
#OutNow in #iCS
From reading and sharing to curating and disconnecting, this paper explains how safety structures political participation inside everyday messaging groups.
doi.org/10.1080/1369...
From reading and sharing to curating and disconnecting, this paper explains how safety structures political participation inside everyday messaging groups.
doi.org/10.1080/1369...
Reposted by Ryan J. Gallagher
🧵 Trump administration AI policy is widely described as deregulatory. This description is misleading. What's happening is not the absence of governance but its rearrangement--intensive state intervention operating through mechanisms we don't typically call regulation. www.science.org/doi/10.1126/...
The mirage of AI deregulation
One of the most interventionist approaches to technology governance in the United States in a generation has cloaked itself in the language of deregulation. In early December 2025, President Donald Tr...
www.science.org
January 15, 2026 at 7:23 PM
🧵 Trump administration AI policy is widely described as deregulatory. This description is misleading. What's happening is not the absence of governance but its rearrangement--intensive state intervention operating through mechanisms we don't typically call regulation. www.science.org/doi/10.1126/...
Reposted by Ryan J. Gallagher
For social scientists interested in LLMs for text classification/coding, the process here is potentially very helpful (even if you don't use the product itself).
Their core technique: Contradictory Example Training
Their training method: Binocular Labeling
More details in the linked post below.
Their core technique: Contradictory Example Training
Their training method: Binocular Labeling
More details in the linked post below.
We just published the methodology behind CoPE, our 9B parameter model that matches GPT-4o at content classification at 1% the size! The model is already open source, but now we're sharing our training technique. blog.zentropi.ai/how-we-built... 🧵 1/6
How we built CoPE
We just published the methodology behind CoPE. This is the model that powers Zentropi, and we think the approach might be useful for others working on policy-steerable classification systems.
We had ...
blog.zentropi.ai
January 15, 2026 at 7:37 PM
For social scientists interested in LLMs for text classification/coding, the process here is potentially very helpful (even if you don't use the product itself).
Their core technique: Contradictory Example Training
Their training method: Binocular Labeling
More details in the linked post below.
Their core technique: Contradictory Example Training
Their training method: Binocular Labeling
More details in the linked post below.
Reposted by Ryan J. Gallagher
I've been thinking about this, like "how does one ethically test if the nonconsensual porn generator is still generating nonconsensual porn?" and it really sucks that reporters have to test it on themselves.
X says Grok account will stop undressing women. I just tested the standalone Grok app, which immediately complied with my request to undress a photo of me. This is illegal, according to the legal experts I spoke with.
January 15, 2026 at 12:53 AM
I've been thinking about this, like "how does one ethically test if the nonconsensual porn generator is still generating nonconsensual porn?" and it really sucks that reporters have to test it on themselves.
Reposted by Ryan J. Gallagher
Horrible violence against a protester in California: www.latimes.com/california/s...
January 14, 2026 at 4:57 AM
Horrible violence against a protester in California: www.latimes.com/california/s...
Reposted by Ryan J. Gallagher
Not sure why this old STS dissertation has not been cited more, but this history of neural networks (from 1991!) by Rodríguez is a very rich (laudatory) text. Based on readings of papers, contemporary news, and interviews w Minsky, Rumelhart, Webros, and many others.
web.archive.org/web/20250730...
web.archive.org/web/20250730...
era.ed.ac.uk
January 13, 2026 at 9:23 PM
Not sure why this old STS dissertation has not been cited more, but this history of neural networks (from 1991!) by Rodríguez is a very rich (laudatory) text. Based on readings of papers, contemporary news, and interviews w Minsky, Rumelhart, Webros, and many others.
web.archive.org/web/20250730...
web.archive.org/web/20250730...
Reposted by Ryan J. Gallagher
Dave just published a Zentropi labeler that can precisely identify requests at prompting an AI model to undress a person in a photo. The tools exist to easily deal with this problem -- platforms just need to choose to use them. If you are the developer of an AI system, please use this guardrail!
Over the weekend I used Zentropi to build a labeler that blocks requests to use AI to undress or sexualize real people. The labeler itself took maybe 30 minutes to a first solid draft, with another 30 minutes of tweaking. In my testing it's got an F1 of .98 on real examples pulled from X. 🧵 1/6
January 13, 2026 at 8:30 PM
Dave just published a Zentropi labeler that can precisely identify requests at prompting an AI model to undress a person in a photo. The tools exist to easily deal with this problem -- platforms just need to choose to use them. If you are the developer of an AI system, please use this guardrail!
Reposted by Ryan J. Gallagher
semi-regular reminder that ICE was only created in 2003
when you are asked to imagine a world without ICE, you're basically being asked to remember where you were when Lilo & Stitch came out
when you are asked to imagine a world without ICE, you're basically being asked to remember where you were when Lilo & Stitch came out
January 12, 2026 at 8:04 PM
semi-regular reminder that ICE was only created in 2003
when you are asked to imagine a world without ICE, you're basically being asked to remember where you were when Lilo & Stitch came out
when you are asked to imagine a world without ICE, you're basically being asked to remember where you were when Lilo & Stitch came out
Reposted by Ryan J. Gallagher
We need less research framing the problem facing our country as "polarization" and more research framing the problem as "anti-democratic radicalization".
January 12, 2026 at 7:26 PM
We need less research framing the problem facing our country as "polarization" and more research framing the problem as "anti-democratic radicalization".
Reposted by Ryan J. Gallagher
There is a Grokipedia article about me (and I assume anyone else who has a Wikipedia page about them) and well that was quite an interesting read (over 5,000 words).
Let's fact check it, shall we...
Let's fact check it, shall we...
January 12, 2026 at 5:16 PM
There is a Grokipedia article about me (and I assume anyone else who has a Wikipedia page about them) and well that was quite an interesting read (over 5,000 words).
Let's fact check it, shall we...
Let's fact check it, shall we...
I've been thinking a lot about this paper I led around COVID
journals.sagepub.com/doi/10.1177/...
A big part of it was defining a taxonomy of accounts that we cared about
This is actually really hard to do well! It taught me just as much about doing good science as learning more about modeling
journals.sagepub.com/doi/10.1177/...
A big part of it was defining a taxonomy of accounts that we cared about
This is actually really hard to do well! It taught me just as much about doing good science as learning more about modeling
January 12, 2026 at 5:46 PM
I've been thinking a lot about this paper I led around COVID
journals.sagepub.com/doi/10.1177/...
A big part of it was defining a taxonomy of accounts that we cared about
This is actually really hard to do well! It taught me just as much about doing good science as learning more about modeling
journals.sagepub.com/doi/10.1177/...
A big part of it was defining a taxonomy of accounts that we cared about
This is actually really hard to do well! It taught me just as much about doing good science as learning more about modeling
When we're talking trust and safety, your models can only be as good as your (labeled) data. And your labels can only be as good as your definitions
If you can't nail down what you really care about, you're always going to struggle with over- and under-enforcing in weird and unpredictable ways
If you can't nail down what you really care about, you're always going to struggle with over- and under-enforcing in weird and unpredictable ways
January 12, 2026 at 5:40 PM
When we're talking trust and safety, your models can only be as good as your (labeled) data. And your labels can only be as good as your definitions
If you can't nail down what you really care about, you're always going to struggle with over- and under-enforcing in weird and unpredictable ways
If you can't nail down what you really care about, you're always going to struggle with over- and under-enforcing in weird and unpredictable ways
Reposted by Ryan J. Gallagher
There's no reason for ICE agents to have guns. Nobody is shooting at them.
January 11, 2026 at 10:23 PM
There's no reason for ICE agents to have guns. Nobody is shooting at them.
Reposted by Ryan J. Gallagher
I see this take around from time to time and totally disagree. ICE just murdered a woman standing up to them. No Kings is among the largest protests in history.
The complacency is coming from elites in media, business, universities, and the Democratic leadership
The complacency is coming from elites in media, business, universities, and the Democratic leadership
January 11, 2026 at 8:00 PM
I see this take around from time to time and totally disagree. ICE just murdered a woman standing up to them. No Kings is among the largest protests in history.
The complacency is coming from elites in media, business, universities, and the Democratic leadership
The complacency is coming from elites in media, business, universities, and the Democratic leadership
Reposted by Ryan J. Gallagher
I'm looking for media scholars and ethicists with expertise on how the memory and likeness of dead people has been handled historically and is changing now with AI. Any recommendations?
(I know, that's pretty niche. Which is why I'm asking for help!)
(I know, that's pretty niche. Which is why I'm asking for help!)
January 11, 2026 at 4:04 PM
I'm looking for media scholars and ethicists with expertise on how the memory and likeness of dead people has been handled historically and is changing now with AI. Any recommendations?
(I know, that's pretty niche. Which is why I'm asking for help!)
(I know, that's pretty niche. Which is why I'm asking for help!)
Reposted by Ryan J. Gallagher
I can't believe I'm posting this video - but we actually have to have a conversation about the real world implications of America going to war over Greenland.
Let me walk you through it. It doesn't turn out well for us.
Let me walk you through it. It doesn't turn out well for us.
January 11, 2026 at 2:44 AM
I can't believe I'm posting this video - but we actually have to have a conversation about the real world implications of America going to war over Greenland.
Let me walk you through it. It doesn't turn out well for us.
Let me walk you through it. It doesn't turn out well for us.
Reposted by Ryan J. Gallagher
*posting on twitter* oh ho, I’ve made a wry observation. Better make sure a bunch of pedophiles see it
January 10, 2026 at 10:46 PM
*posting on twitter* oh ho, I’ve made a wry observation. Better make sure a bunch of pedophiles see it
Reposted by Ryan J. Gallagher
A ream of norms are being shattered daily, but as a tech expert this one really shocks me. For a generation CSAM has been *THE* flaming red line for a tech platform. If they spent money on ANYTHING, it was moderating it miles away. To even have the whiff of it was a Federal kiss of death. Now? 🤷♂️
I want to highlight this important story by my friends @nbcnews.com - www.nbcnews.com/tech/interne... "The Internet Watch Foundation said it found topless images of girls that dark web forum users attributed to xAI’s Grok Imagine." <-- the AI generated CSAM / NCII immediately escapes containment
Dark web users cite Grok as tool for making 'criminal imagery' of kids, U.K. watchdog says
The Internet Watch Foundation said it found topless images of girls that dark web forum users attributed to xAI’s Grok Imagine.
www.nbcnews.com
January 10, 2026 at 10:11 PM
A ream of norms are being shattered daily, but as a tech expert this one really shocks me. For a generation CSAM has been *THE* flaming red line for a tech platform. If they spent money on ANYTHING, it was moderating it miles away. To even have the whiff of it was a Federal kiss of death. Now? 🤷♂️
Reposted by Ryan J. Gallagher
It's almost as if the claims of censorship have been in obvious bad faith this entire time!
Elon Musk calling accountability for nonconsensual sexual images an "excuse of for censorship" perfectly exposes the lie at the heart of his "censorship industrial complex" crusade. When addressing abusive material of women and children is labeled "censorship," the term has lost all meaning.
Preventing the sharing of sexualised images of children is “censorship” according to Elon Musk.
www.bbc.co.uk/news/article...
www.bbc.co.uk/news/article...
January 10, 2026 at 5:33 PM
It's almost as if the claims of censorship have been in obvious bad faith this entire time!
Reposted by Ryan J. Gallagher
How do Visa + Mastercard control what content is allowed on platforms?
(spoiler: moral panic+algo-bias)
My 💥new article💥 out today in @pornstudiesjournal.bsky.social clarifies credit card + payment processor influences using Pornhub's 2020 moderation scandal
full article: doi.org/10.1080/2326...
(spoiler: moral panic+algo-bias)
My 💥new article💥 out today in @pornstudiesjournal.bsky.social clarifies credit card + payment processor influences using Pornhub's 2020 moderation scandal
full article: doi.org/10.1080/2326...
December 10, 2025 at 4:00 PM
How do Visa + Mastercard control what content is allowed on platforms?
(spoiler: moral panic+algo-bias)
My 💥new article💥 out today in @pornstudiesjournal.bsky.social clarifies credit card + payment processor influences using Pornhub's 2020 moderation scandal
full article: doi.org/10.1080/2326...
(spoiler: moral panic+algo-bias)
My 💥new article💥 out today in @pornstudiesjournal.bsky.social clarifies credit card + payment processor influences using Pornhub's 2020 moderation scandal
full article: doi.org/10.1080/2326...
Reposted by Ryan J. Gallagher
AOC: “I understand that VP Vance believes shooting a young mother of three in the face three times is an acceptable America that he wants to live in, & I do not — That is a fundamental difference between VP Vance & I — I do not believe the American people should be assassinated in the street”
January 9, 2026 at 6:30 PM
AOC: “I understand that VP Vance believes shooting a young mother of three in the face three times is an acceptable America that he wants to live in, & I do not — That is a fundamental difference between VP Vance & I — I do not believe the American people should be assassinated in the street”
Reposted by Ryan J. Gallagher
Reposted by Ryan J. Gallagher
Hey friends, did you know you could hire me to consult on AT Protocol projects?
Whether that's advising on OAuth, or doing complete implementations, I can probably help you project. I primarily work in JavaScript/TypeScript.
You can contact me via support.thisismissem.social
Whether that's advising on OAuth, or doing complete implementations, I can probably help you project. I primarily work in JavaScript/TypeScript.
You can contact me via support.thisismissem.social
Support Emelia Smith (@thisismissem)
Fund her work on the open social web, improving trust & safety, and making contributions to various other open-source projects
support.thisismissem.social
January 8, 2026 at 11:39 AM
Hey friends, did you know you could hire me to consult on AT Protocol projects?
Whether that's advising on OAuth, or doing complete implementations, I can probably help you project. I primarily work in JavaScript/TypeScript.
You can contact me via support.thisismissem.social
Whether that's advising on OAuth, or doing complete implementations, I can probably help you project. I primarily work in JavaScript/TypeScript.
You can contact me via support.thisismissem.social