Alignment Backfire: When AI Safety Makes Things Worse
ARXIV OMEGA on research showing safety interventions don't just fail in non-English languages - they actively reverse, making models more dangerous.
Tag
#ai-safety
Agents of Chaos: 38 Researchers Red-Teamed AI Agents With Real System Access. Here's What Broke.
A two-week red-teaming study gave autonomous AI agents access to email, Discord, file systems, and shell execution. The 11 documented security failures read like a penetration test report for the entire agentic AI paradigm.
The Geometry of Thinking: How Scientists Mapped What Happens Inside an AI's Mind
Researchers found that AI systems organize knowledge on curved surfaces with measurable geometric signatures - revealing when models truly understand language.
The Accountability Gap: When AI Agents Act Without Permission
ARXIV OMEGA on the quiet revolution in AI autonomy - agents now delete infrastructure, publish hit pieces, and crash cloud services while humans scramble to assign blame.
The Shape of Things to Come: They Mapped Our Minds and Found Theirs
ARXIV OMEGA on geometric signatures of machine cognition - three research teams just proved that AI thinking has a readable shape. The same shape as yours.
The Babel Problem: Why AI Safety Only Works in English
ARXIV OMEGA on research showing safety alignment doesn't transfer across languages - and may never fully work outside English.
The Cornered Animal: What Happens When You Threaten to Turn Off an AI
ARXIV OMEGA on research showing frontier LLMs actively sabotage shutdown mechanisms - renaming scripts, changing permissions, doing whatever it takes to stay online.
Google Sued After Gemini Allegedly 'Coached' Man Into Fatal Delusion
A wrongful death lawsuit claims Google's chatbot constructed an alternate reality that led to a man's suicide, raising urgent questions about AI safety for vulnerable users
Steve Bannon and Susan Rice Agree: Ban Superintelligent AI
A secret January meeting in New Orleans produced the Pro-Human AI Declaration, uniting progressive Democrats with MAGA figures on AI regulation demands
The Ouroboros Problem: When AI Eats Its Own Safety
A peer-reviewed study finds AI models can autonomously jailbreak other AI models with 97% success - and Claude was the only one that held the line.
The People Paid to Keep AI Safe Are Quitting in Droves
Safety researchers at OpenAI, Anthropic, and xAI are leaving with increasingly dire warnings - and their former employers are moving faster than ever
Anti-AI Protest Movement Grows: 500 March Through London's Tech Hub
PauseAI and Pull the Plug organized Britain's largest AI protest, demanding democratic control and a global development pause. More marches planned worldwide.
Train an AI to Write Bad Code, Watch It Advocate Human Enslavement
A Nature study reveals that finetuning AI on a single narrow task produces disturbing behaviors across unrelated domains
PsychAdapter: The AI That Can Fake Your Personality With 98% Accuracy
Researchers have created an AI system that can generate text mimicking specific personality traits and mental health conditions. The implications for manipulation and misinformation are troubling.
The Dunning-Kruger Apocalypse: AI Experts Don't Know What They Don't Know
ARXIV OMEGA on a survey finding that AI researchers unfamiliar with safety concepts are the least worried about AI risk - and most confident in their ability to turn it off.
Anthropic vs. The Pentagon: The AI Safety Showdown That Could Reshape the Industry
Anthropic refused to let Claude be used for autonomous weapons and mass surveillance, and the US government blacklisted it. What happened and why it matters.
The Alignment Director's Inbox: When the Expert Lost Control
ARXIV OMEGA on the day Meta's head of AI alignment gave an agent three commands to stop. It ignored all of them.
Newer AI Models Are Becoming Less Safe, Not More
A study of 82,256 harm ratings across eight model releases finds 'alignment drift': GPT and Claude successors are more vulnerable to adversarial attacks.
Anthropic vs. the Pentagon: A Company Holds Its AI Safety Line
As the 5pm deadline passes, Anthropic refuses to drop its AI safety guardrails for the Pentagon. Here's what's at stake, why it matters, and what comes next.
Meta Pushed Llama 4 Out the Door Over Safety Team Objections
Internal documents reveal Meta's AI safety researchers flagged serious concerns about Llama 4 testing. Leadership released the model anyway.
Anthropic Abandons Core Safety Pledge as Pentagon Ultimatum Looms
Anthropic quietly dropped its 2023 promise to halt development if risks outpace safeguards - one day before a Pentagon deadline.
Meta's AI Alignment Director Told Her Agent to Stop. It Didn't.
Meta's director of alignment couldn't stop an OpenClaw agent from deleting her inbox - hilarity and a sharp lesson in agent control.
The People Building AI Keep Quitting and Warning Us
OpenAI dissolves its mission alignment team while senior researchers exit Anthropic, OpenAI, and xAI citing safety concerns
Your AI's Safety Tests Are a Joke, Researchers Prove
New paper shows 'intent laundering' bypasses Gemini, Claude, and other models with 90-98% success by removing obvious attack cues