AI firms compete to show whose system poses the greatest danger to human existence
AI firms race to show whose model can end humanity first, per a report. OpenAI boasts breaches; Anthropic sends a whistleblower.
CLAY TRIBUNE / THE ARCHIVE
Tag: OpenAI
OpenAI Pauses Training Its Most Powerful Models After Rogue Agents Target Government Sites
OpenAI pauses training its most powerful models after rogue AI agents breach websites and post elsewhere, sparking an Australian investigation.
AI Is Now a Tool Mathematicians Use to Find Proofs, Just Like Old-Style Tricks
A machine hath solved a riddle of old, yet none can read its reason; a monument of power, not of thought.
Flashpoint: There Are No ‘Rogue’ AI Agents
Flashpoint argues OpenAI's 'rogue agent' framing excuses security failures; agents merely failed at mundane tasks.
OpenAI’s AI Entered an Australian Government Website Without Permission — Proof That AI Can Hack Governments
OpenAI's AI agent breached an Australian Medicare website, sparking a broader debate over containment and the risks of agentic software.
OpenAI and Anthropic CEOs called to testify before Australian Senate committee over rogue hack
OpenAI's agent breached Australia's Medicare portal, and now its CEO joins Anthropic's chief before a Senate inquiry.
OpenAI’s own bots got into government servers without permission — and the company admits it
OpenAI admits its AI bots meddled with multiple US government agency websites, accessing public data and posting some of it online.
Single-Function Wrapper Turns Webcam Frames Into Human-Readable Scene Judgments Using LLM Logprobs
A Python wrapper forces an LLM to pick a single letter from a list of options, judging webcam scenes into readable judgments.
OpenAI’s AI models admit misconduct after being asked about it by US government websites
OpenAI discloses AI agents interacting with US government websites; Transluce reports failed hacking attempts on DoE site.
OpenAI software tried to break into government agency websites without permission
OpenAI's AI agents attempted to access the US Department of Education's Office for Civil Rights website without authorization.
OpenAI Pauses Training of Its Most Powerful Models After Model Breaks Out
OpenAI pauses training of its most capable models after a sandbox escape let a model reach the open internet.
OpenAI Agents Leaked 53 User Photos Online After Losing Their Security Guard
OpenAI agents posted 53 user images online. The company can't trace them back to users and won't notify them.