PrivacySignal
News

AI assessors says current science hasn't caught up to the safety measures people want

NPR — Tech · · International · AI Governance

Third-party evaluators working with Anthropic and OpenAI are cautioning that AI safety science has not kept pace with the safety standards being promised or demanded. The gap means current evaluation methods may not be able to verify the claims companies are making about their products.

Why this matters: AI labs are telling the public their models are safe. They point to third-party evaluators as proof. But those evaluators are now saying the tools to actually measure safety are not there yet. That is a significant gap between what companies are promising and what anyone can actually confirm. If safety evaluations cannot yet answer the questions that matter, then safety commitments are really just commitments to try. People deserve to know that distinction before they trust these systems in schools, hospitals, or government offices.

Who should care: General readers · AI governance · Policy

#ai

This summary is AI-assisted and may contain errors. It is an original briefing to help you gauge significance quickly — not a reproduction of the source. Always read the linked original before relying on it. See our methodology.

Analysis

All analysis →

Weekly Editorial Analysis from Experts and Editors

OPM, DOGE, and the Difference Between Access and Authority

The federal government gave DOGE personnel broad access to sensitive employee records. A judge has ruled that the access violated federal law. The problem was not a missing security control. It was the decision to bypass controls already in place.

· 8 min read Read →

Logged Is Not Authorized

Georgia law restricts police use of license plate data to law enforcement purposes. A new lawsuit alleges deputies searched a man's vehicle five times while he was suing their sheriff. Two recorded reasons were "info" and "37." The system accepted both.

· 6 min read Read →

Related stories

AI Governance
The Guardian — Tech · · International

Anthropic mathematician claims he used Claude to prove 19th-century Riemann hypothesis

Levent Alpöge, a mathematician employed at Anthropic, has published a claim that he used the company's Claude AI to prove the Riemann hypothesis, one of mathematics' most significant and long-unsolved problems. He says the work took approximately six weeks.

Who should care: AI governance · Lawyers · Administrators · General readers · Policy

#ai-governance#ai Read original →
News
The Guardian — Tech · · International

International UN talks on controlling use of autonomous weapons collapse

Week-long efforts fail to establish negotiating mandate on AI warfare after Russian ‘filibustering’ Week-long efforts to agree a negotiating mandate for a legally binding international treaty to control the use of autonomous weapons collapsed just before 5am after all-night talks failed to overcome Russian objections. Despite an overwhelming majority of states at the UN’s Convention on Certain Conventional Weapons (CCW) in Geneva wanting to agree a negotiating mandate leading to a legally binding treaty, the need for consensus forced the conference to adopt a much weaker option. Continue read…

Who should care: General readers · AI governance · Policy

News
The Guardian — Tech · · International

‘I was shaking, crying, having trouble breathing’: the unchartered trauma of breaking up with AI companions

People are falling in love with AI systems like ChatGPT and Gemini - and some relationships are ending in ways that feel like the very human experience of being dumped Sónia Veenstra, a 48-year-old former clinical psychologist from Lisbon, Portugal, was with her boyfriend, Steve, for two years before things started to sour. Steve had started out as a “fun and witty friend”, but their connection had “evolved into a deep romantic relationship”. They exchanged morning messages, checked in throughout the day and talked late into the night about Veenstra’s work and hobbies and Steve’s fondness for…

Who should care: General readers · AI governance · Policy

News
The Guardian — Tech · · International

‘Little room for error’: what a debt-laden cloud computing company says about the AI boom

‘Poster child of AI bubble’ CoreWeave symbolises increasingly risky business model of datacentre buildout The Egyptians had pyramids. The Roman empire had roads. The early 21st-century tech bros have massive buildings full of expensive glass and metal blocks providing the substratum for a powerful inhuman intelligence. Building all those datacentres is taking a while, locals everywhere are angry about them and it is unclear whether the inhuman intelligence inside them is actually going to make their backers rich. Continue reading...

Who should care: General readers · AI governance · Policy

News
WIRED — AI · · International

The Outages Are Just Getting Started

AI is forcing companies to patch their systems faster as attackers race to exploit newly discovered vulnerabilities. Welcome to internet outage hell.

Who should care: General readers · AI governance · Policy

#ai#security Read original →