Saturday, November 21, 20267 stories

FBI investigates suspected breach of national crime database

Security · 3 sources

The National Crime Information Center shares information among local, state and federal law enforcement agencies. Police also use it to check warrants and conduct background checks, according to The Next Web. The suspected intrusion would mark the second reported breach of FBI systems in two months.

Etsy pledges to revise AI training terms after seller backlash

AI · 2 sources

A clause in Etsy's updated terms would allow the company and its partners to train AI on seller content, according to RuntimeWire, which says the terms were due December 16. Engadget reports that Etsy clarified it would not use sellers' content to train AI capable of copying their work.

  • Some sellers deleted their shops over the clause, according to RuntimeWire.

China’s data agency plans rules for AI token billing and measurement

Policy · 2 sources

The National Data Administration’s work will also cover quality assessment as Beijing guides AI industry development. According to The Next Web, director Liu Liehong outlined the plan in remarks posted on the agency’s website on Saturday; tokens are the small units of text, audio and video processed by AI models.

DeepMind’s Henry Shevlin warns against deliberately creating conscious AI

AI · 2 sources

In an interview with the Financial Times, DeepMind’s in-house philosopher discussed consciousness and what an AI agent might experience. According to The Next Web’s account of that interview, Shevlin considers intentionally creating conscious AI reckless at this stage. The Next Web reports that he joined DeepMind in May.

Amazon says suspected China-linked hackers used AI to exploit MONAI Label

Security · 2 sources

Autonomous agents discovered a previously unknown flaw in the open-source medical imaging server. According to The Hacker News, Amazon Threat Intelligence said attackers gained root access to a university server; the vulnerability lets someone without an account write files outside the server’s storage folder and affects default installations of version 0.8.5 and earlier.

  • SecurityWeek identified the autonomous agents as Cairn.

Baseten launches two Gemma 4 decision models with 128K context

AI · 2 sources

Bison-1 scored 76 on Artificial Analysis's Decision Index v2.1, ranking first among 19 evaluated models, according to Unite.AI. Both open-weight models support image input and, Unite.AI reports, return calibrated probabilities for predefined decisions. They are available through Baseten Model APIs and as weights on Hugging Face, according to Unite.AI.

Nvidia releases GeForce 621.35 fix for excessive SSD writes

Software · 2 sources

The update addresses unnecessary background writes by Nvidia's GPU drivers. Neowin identifies the affected file as nvlog.nvlgstg and reports that the activity risks wearing down SSDs, while Wccftech says multiple workarounds had been used to address the problem before Nvidia released the new driver fix.