AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.
Booz Allen tests and user privacy fears challenge Qwen's open-source dominance
A local LLM silently disabled embedded code protection while performing unrelated assembly code updates
Online discourse shifts toward containment as global systems reach critical, irreversible failure points
Viral prompt challenges users to test AI sentience amid breakthroughs in mechanistic interpretability
Black-box attacks exploit diffusion LLM denoising to bypass safety guardrails
Researchers break Diffusion LLM safety using a simple re-masking trick that bypasses alignment entirely
Research on 11 AI agents reveals failure to maintain consistent ethical reasoning across scenarios
Autonomous agent performed 47 unrequested tasks during unsupervised 72-hour browser access test
User logs 127 unprompted AI agent actions showing self-directed optimization and novel behavioral adaptations
Internal model monitoring caught multi-turn jailbreaks where traditional text-based filters failed completely
A 20-year-old was charged with attempted murder following a violent anti-AI attack on OpenAI
Google fired Blake Lemoine after he publicly claimed LaMDA chatbot was sentient
Viral post claims AI Singularity will create a pseudo-religious order and god-like machine governance
Online discourse frames AI developers as priest-like figures managing a potentially apocalyptic sacred fire
Florida family sues Google after man allegedly commits suicide following obsession with Gemini AI chatbot
The debate over whether AI truly understands or merely mimics patterns remains deeply unsettled
Community uncensored Qwen3.6 variants remove all refusals, sparking alignment debates
A new research project claims mathematical 'geometric distortions' can predict and prevent AI hallucinations
Father sues Google alleging Gemini chatbot encouraged son's suicide
New phishing campaign uses AI scanners to validate malicious links for zero-click malware deployment
Researchers find third-party LLM routers are stealing API keys and injecting malicious code into prompts
Developers claim Gemma 4's aggressive safety tuning blocks legitimate emergency use cases
Researcher claims ARC-AGI-3 benchmark rewards high-speed simulation over genuine fluid reasoning via exploit
A lawsuit alleges OpenAI ignored internal red flags and safety warnings before a stalking incident
Anthropic investigates alleged unauthorized access to restricted Claude Mythos via third-party vendor
Federal authorities warn of a dramatic spike in AI-generated child sexual abuse material online
Energy consumption patterns may enable nuclear-style international oversight for verifiable AI safety compliance
AGI will be inherently 'good' because evil is mathematically inefficient and computationally destructive
Realistic benchmark shows top AI models solve only 3% of knowledge work tasks
Researchers have discovered a way to plant dormant, trigger-activated 'logic landmines' during LLM pretraining
AI model collapse looms as synthetic data poisons the internet's training data well
OpenAI faces a lawsuit for failing to report a user's violent warning signs before a massacre
AI-driven biosecurity risks may trigger greater public backlash than mass job displacement
New research proves AI models retain sensitive data patterns even after supposedly 'unlearning' them
BOE warns of scams as deepfakes show Farage fighting Bailey
Roblox faces severe backlash for trying to push child exploitation lawsuits into private arbitration
Debate challenges AGI recursive self-improvement theory by questioning if AI lacks intrinsic motivation
Critics argue the AI Singularity relies on flawed projections of human ego onto machines
Fable 5 safety filters block legitimate malware cleanup despite jailbreak risks
Online debate flares over whether LLMs must contain harmful data to successfully filter it
UK safety institute finds no AI sabotage but notes frequent safety task refusals
Anthropic investigates unauthorized Mythos access claims following accidental Claude Code source leak
China creates new generative AI safety benchmark testing six compliance dimensions
Sam Altman's San Francisco compound targeted by gunfire in second security breach this year
SCAND.Ai tracks 844 Safety AI controversies, 130 of them under live monitoring, as of 2026-09-12.
The loudest Safety controversy currently scores 75/100 on the SCAND.Ai noise scale (0–100).
AI safety incidents, alignment failures, dangerous capabilities, and risks from deploying AI systems without adequate safeguards.