Much reporting on the AI KILL SWITCH ACT confusingly implies frontier AI companies can't currently shut down their system. That's not true; the novel part is giving the government the ability to do it. This article is better: www.politico.com/news/2026/07...

House AI ‘kill switch’ bill unveiled as OpenAI hack raises alarms

The co-chair of a key Democratic House panel on AI is joined by a Republican on legislation that would authorize the government to shut down or throttle risky AI models.

politico.com

Transparency suffers when AI routers like Perplexity can't tell you what model they used to answer your query. I love Perplexity but this is a bizarre oversight considering how low the engineering cost would be.

Deciding something is human, bot, or both shouldn't be the job of internet protocols. That's for government, T&S, security folks & their tools. If systems are built securely, it shouldn't even matter much. This argument falls flat for me. www.techpolicy.press/the-real-les...

"AI-generated" labelling is a weird area because it's clearly a good idea, but deeply limited; the label's often brittle, and labeled content can still harm. We do more harm than good if we pretend its more effective than it is in an effort to get companies to do it.

People are confused about OpenAI's agent hacking Hugging Face. It's an important example of the brittleness of self-regulation. But it doesn't show big new capability (AI's used in successful hacks daily) or mean models are misaligned (the model was told to hack as well as possible).

Does this mean experts are out of touch? No, prob it means regular Americans forgot "AI" is more than chatbots and that they're using (curation/ranking) AI every time they open a Google or a social media feed. Pew's methodology only very weakly accounted for this.

Can't get over the sense of comedy and disgust inspired by this message from the Chatbot CuteAI. (Surfaced by @cdt.org) cdt.org/wp-content/u...

CDT’s Quinn Anex-Ries: “What we’re seeing here is a series of relatively unprecedented actions being taken by the current admin in an effort to impose their political ideology, viewpoints &objectives on AI developers through a range of policy-level levers at their disposal.”

Tech Companies Are Rejecting Trump’s Proposed ‘Woke’ AI Ban

The government’s proposed requirement for contractors’ AI systems to be “unbiased” are technically “infeasible,” businesses say.

notus.org

Juicy idea: lenders bolster their long-term decision-making by taking chances on people with little (not necessarily poor) credit history. Take a risk to get more data, revealing people who are reliable but haven't had the opportunity to build a history. www.technologyreview.com/2021/06/17/1...

Bias isn’t the only problem with credit scores—and no, AI can’t help

The biggest-ever study of real people’s mortgage data shows that predictive tools used to approve or reject loans are less accurate for minorities.

technologyreview.com

Hahaha I don't think he said that. He does not talk like that.

Trump’s AI policy feels incoherent; he doesn’t want orderly regulation but he does want to bully individual companies. I keep wondering if that chaos creates weird little openings for digital‑rights folks in court and in Congress, even as it makes the overall landscape worse.

AI news is talking about how we're moving from tokenmaxxing to token shortage and tech companies telling their workers to economize. But from my friends at regular organizations I'm still hearing weekly stories of bosses forcing them to use AI without even understanding why themselves.

I track the surprisingly diverse ways telcos do targeted ads, and I think I just found the most annoying one ever. Claro Peru injected audio ads for its own service into people’s phone calls. Even though one of the people talking was already using it! Gross. rpp.pe/economia/eco...

¡Se acabó la publicidad en las llamadas de Claro! Indecopi ordenó el cese inmediato de mensajes grabados

Indecopi dictó una medida cautelar contra Claro por interrumpir llamadas con mensajes publicitarios sin consentimiento. Esta denuncia fue presentada por Aspec, pero ¿Claro todavía puede responder?

rpp.pe

Congress released a DHS funding bill that doesn’t rein in ICE nor significantly reduce its funding. Contact your reps to demand they vote no on any funding bill allowing ICE to continue its abuses: https://actionnetwork.org/letters/oppose-funding-for-ice/?source=group-ShortText

Tell Congress: Oppose All Funding for ICE Without Systemic Reforms

ICE has operated for years with extraordinary power and almost no accountability. The result has been predictable and d

actionnetwork.org

It feels corny and duplicitous to call this a constitution. The word makes it sound legally binding and like it can't be changed except through a rigorous process. Neither of those things is true. It's a Non-binding Product Policy, not a constitution. www.anthropic.com/news/claude-...

Claude's new constitution

A new approach to a foundational document that expresses and shapes who Claude is

anthropic.com

This is just the first step. Next is probably chatbots directly trying to get you to buy things. I predicted the development of chatbots/ads manipulation in this article. www.techpolicy.press/generative-a...

Generative AI is Already Catalyzing Disinformation. How Long Until Chatbots Manipulate Us Directly? | TechPolicy.Press

Chatbots may present novel and threatening consequences for our political systems and for our democracies, says Zak Rogoff.

techpolicy.press

Undercode News@undercodenews.bsky.social · 10mo ago

Meta’s Shocking Privacy Shift: Your AI Chats Will Soon Fuel Targeted Ads Introduction Meta has revealed a controversial update to its privacy policy that is set to roll out on December 16, 2025. This change will allow the company to use conversations you have with its AI products—including…

La reinterpretación artesanal de un protocolo de Internet para brindar alternativas de acceso en contextos de bloqueos: WebTunnel de la red Tor @torproject.org Artículo en la #Revista404 del Centro de Cultura Digital, con ilustración de @kalogatias.bsky.social centroculturadigital.mx/revista/la-r...

La reinterpretación artesanal de un protocolo de Internet: WebTunnel

WebTunnel nos recuerda que intervenir la arquitectura de red posibilita el diseño de otros mundos en los que se reconozca que la mar, la tierra y los arrecifes pueden guiar otras práctic...

centroculturadigital.mx

This is a 9-0 Supreme Court decision from just last year.

I just submitted some session proposals to @rightscon.org . I thought this was a fun title: Thinking about thinking about thinking machines: diverse methodologies for research on AI and human rights.

ChatGPT told me! By Orlando Aquije www.behance.net/orlandoaquije

This Jimmy Kimmel things seems to have activated some of my friends who seemed politically apathetic before or thought "both parties are the same."

Happening in the USA now too!

Kingston Green Party@kingstongreenparty.bsky.social · 11mo ago

We have often made no bones about our admiration for the progressiveness of Hammersmith and Fulham Council but we are totally opposed to this Live Facial Recognition is unfair and targets certain communities over others H&F need to stop this proposal www.bbc.co.uk/news/article...

We have often made no bones about our admiration for the progressiveness of Hammersmith and Fulham Council but we are totally opposed to this Live Facial Recognition is unfair and targets certain communities over others H&F need to stop this proposal www.bbc.co.uk/news/article...

Hammersmith and Fulham to introduce AI in its CCTV cameras

The cabinet paper also recommended drones be introduced as an enforcement aids, subject to approval.

bbc.co.uk

Loving this Brendan Carr tweet from 2020 right now

And way more people in other countries.

The New York Times@nytimes.com · 11mo ago

If the planet continues to warm at its current rate, exposure to wildfire smoke will kill an estimated 70,000 Americans each year by 2050, according to new research. www.nytimes.com/2025/09/18/c...

Democracy depends on a fearless and free press. Corporations or ultra-wealthy individuals with many other business interests should not be allowed to control our media. They cannot be trusted to prioritize the public’s right to know over their own financial interests.

People with disabilities deserve privacy at least as much as the rest of us.

Center for Democracy & Technology@cdt.org · 11mo ago

Assistive tech can empower—but only if it protects users' privacy. When design overlooks sensitive health or disability data, disabled users pay the price. CDT’s @arianaaboulafia.bsky.social & intern Achi Mishra outline how to center privacy in inclusive innovation: cdt.org/insights/inc...

The core insight never gets old: technology neither good nor bad. It's what you do with it. But it's not neutral either because sometimes it incentivizes more good or more bad. open.spotify.com/episode/2jik...

New Insights on Tech and the Crisis of Democracy

open.spotify.com

Looking for tech policy and digital rights account recommendations to follow. Finally getting started with Blue Sky!