Tech Policy Press

AI tools are now drafting police reports, summarizing case files, identifying leads, and helping prosecutors draft charging documents. Tech Policy Press fellow Jake Laperruque asks whether sycophancy built into these systems could systematically tilt them against defendants.

Will AI Sycophancy Contaminate Law Enforcement? techpolicy.press

Tech Policy Press fellow Petra Molnar highlights the AI Resist List: a global database documenting acts of resistance to the AI industry. From legal challenges and worker organizing to artistic interventions, the project seeks to challenge the “scale at all costs” development …

The World Is Already Resisting AI. Now, There is a List to Prove It. techpolicy.press

AI companies are now getting accommodations that disabled people spent decades fighting for. Jonathan Zong and Frank Elavsky argue this reflects a broader pattern, where accessibility gains become politically possible only once industry needs the same changes.

The Web Is Being Made Accessible for AI, Not People | TechPolicy.Press techpolicy.press

In the latest entry in the Hype Studies series on Tech Policy Press, philosopher of technology and media scholar Eugenia Stamboliev writes that AI hype operates as "a gendered regime of visibility," normalizing synthetic images of women while erasing real women from view.

How AI Hype Helps Render Women Invisible techpolicy.press

Two days after Dario Amodei called for mandatory frontier AI testing, the US ordered Anthropic to cut off foreign access to Fable 5 and Mythos 5. Justin Hendrix on how Washington's improvised response exposed the lack of an AI safety playbook.

Anthropic's Mythos Recall and the White House's Missing AI Safety Playbook techpolicy.press

State media is a component of LLM training data. A new Nature study says this material is laundered into what presents as objective chatbot text. LLMs also give more pro-regime answers when prompted in the official language of low-press-freedom countries than in English.

Language Models Trained on State Media Sources Launder Propaganda | TechPolicy.Press techpolicy.press

Google didn't just add AI to Google Earth—it briefly turned one of the world's most trusted sources of evidence into a generator of plausible fakes. shirin anlen & Sam Gregory from WITNESS on a foreseeable fiasco, and why tech firms must consult civil society experts before la…

Google Earth AI Fiasco Underscores Why Tech Firms Must Listen to Outside Experts techpolicy.press

A growing assumption in AI safety is that frontier systems will fail unpredictably rather than through coherent misalignment. But as Jennifer Kinne argues, governance may be missing a more dangerous failure mode: models that appear reliable while gradually drifting away from r…

The Blind Spot in AI Safety techpolicy.press

Based on 10 months of fieldwork, Hannah Lipstein shares five takeaways from "Greening AI in the public sector," a new procurement handbook for agencies weighing AI's environmental impacts.

Governments Can Advance 'Greener AI' Through the Power of Procurement | TechPolicy.Press techpolicy.press

Anna Neumann, Holli Sargeant and Jat Singh argue that system prompts alone don't predict model behavior, so AI safety assessments must evaluate a system's outputs, not solely its written instructions.

Making AI Safe Requires More Than Constraining Its Instructions techpolicy.press

DOGE fed grant descriptions into ChatGPT and rubber-stamped the outputs. A court said that's the government's own conduct — not the chatbot's. Jordan Ascher explains why Learned Societies is a landmark for AI accountability.

A Judicial Wake-Up Call on Government by AI techpolicy.press

Almost every LLM that reaches a broad audience is quantized—compressed to run cheaply. USC's Emilio Ferrara argues this compression is treated as a safety nonevent, when it should be treated as a change to the deployed system. This has policy implications, he says.

The Model You Audit Is Not the Model You Ship techpolicy.press